Pre-trained Models

This page describes how to download pre-trained Fun-ASR-Nano-2512 models.

sherpa-onnx-funasr-nano-int8-2025-12-30 (Chinese, English, Japanese)

This model is converted from Fun-ASR-Nano-2512 using scripts from https://github.com/Wasser1462/FunASR-nano-onnx.

It supports the following 3 languages:

• Chinese

• English

• Japanese

Hint

中文包括 7 种方言（吴语、粤语、闽语、客家话、赣语、湘语、晋语）和 26 种地方口音（河南、山西、湖北、四川、重庆、云南、贵州、广东、广西 及其他 20 多个地区）。

英文和日文涵盖多种地方口音。

此外还支持歌词识别和说唱语音识别。

In the following, we describe how to use it.

Huggingface space

You can visit

https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition

to try this model in your browser.

Hint

You need to first select the language 31 languages (FunASR Nano) and then select the model csukuangfj/sherpa-onnx-funasr-nano-int8-2025-12-30.

Android APKs

Real-time speech recognition Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/vad/apk-asr.html

Please always download the latest version.

Hint

Please search for multi-funasr_nano_int8_2025_12_30.apk in the above page, e.g., sherpa-onnx-1.12.21-arm64-v8a-vad_asr-multi-funasr_nano_int8_2025_12_30.apk.

Hint

For Chinese users, you can also visit https://k2-fsa.github.io/sherpa/onnx/vad/apk-asr-cn.html

Download

Please use the following commands to download it:

cd /path/to/sherpa-onnx

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-funasr-nano-int8-2025-12-30.tar.bz2

# For Chinese users, you can also use
# wget https://modelscope.cn/models/csukuangfj/asr-models/resolve/master/sherpa-onnx-funasr-nano-int8-2025-12-30.tar.bz2

tar xvf sherpa-onnx-funasr-nano-int8-2025-12-30.tar.bz2
rm sherpa-onnx-funasr-nano-int8-2025-12-30.tar.bz2

After downloading, you should find the following files:

ls -lh sherpa-onnx-funasr-nano-int8-2025-12-30/

total 948M
drwxr-xr-x  5 kuangfangjun root    0 Jan  7 19:28 Qwen3-0.6B
-rw-r--r--  1 kuangfangjun root  253 Jan  7 19:33 README.md
-rw-r--r--  1 kuangfangjun root 149M Jan  7 19:33 embedding.int8.onnx
-rw-r--r--  1 kuangfangjun root 227M Jan  7 19:34 encoder_adaptor.int8.onnx
-rw-r--r--  1 kuangfangjun root 573M Jan  7 19:34 llm.int8.onnx
drwxr-xr-x 27 kuangfangjun root    0 Jan  7 19:28 test_wavs

ls -lh sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B/
total 16M
-rw-r--r-- 1 kuangfangjun root 1.6M Jan  7 19:34 merges.txt
-rw-r--r-- 1 kuangfangjun root  11M Jan  7 19:34 tokenizer.json
-rw-r--r-- 1 kuangfangjun root 2.7M Jan  7 19:34 vocab.json

ls -lh sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/
total 9.7M
-rw-r--r-- 1 kuangfangjun root 6.9K Jan  7 19:33 README.md
-rw-r--r-- 1 kuangfangjun root 220K Jan  7 19:33 dia_hunan.wav
-rw-r--r-- 1 kuangfangjun root 253K Jan  7 19:33 dia_minnan.wav
-rw-r--r-- 1 kuangfangjun root 229K Jan  7 19:33 dia_sh.wav
-rw-r--r-- 1 kuangfangjun root 297K Jan  7 19:33 dia_yue.wav
-rw-r--r-- 1 kuangfangjun root 215K Jan  7 19:33 far_2.wav
-rw-r--r-- 1 kuangfangjun root 682K Jan  7 19:33 far_3.wav
-rw-r--r-- 1 kuangfangjun root 284K Jan  7 19:33 far_4.wav
-rw-r--r-- 1 kuangfangjun root 279K Jan  7 19:33 far_5.wav
-rw-r--r-- 1 kuangfangjun root 254K Jan  7 19:33 ja.wav
-rw-r--r-- 1 kuangfangjun root 255K Jan  7 19:33 ja_en_codeswitch.wav
-rw-r--r-- 1 kuangfangjun root 259K Jan  7 19:33 lyrics.wav
-rw-r--r-- 1 kuangfangjun root 431K Jan  7 19:33 lyrics_2.wav
-rw-r--r-- 1 kuangfangjun root 546K Jan  7 19:33 lyrics_3.wav
-rw-r--r-- 1 kuangfangjun root 1.3M Jan  7 19:33 lyrics_en_1.wav
-rw-r--r-- 1 kuangfangjun root 679K Jan  7 19:33 lyrics_en_2.wav
-rw-r--r-- 1 kuangfangjun root 1.7M Jan  7 19:33 lyrics_en_3.wav
-rw-r--r-- 1 kuangfangjun root 331K Jan  7 19:33 noise_en.wav
-rw-r--r-- 1 kuangfangjun root 267K Jan  7 19:33 rag_biochemistry.wav
-rw-r--r-- 1 kuangfangjun root 214K Jan  7 19:33 rag_chemistry.wav
-rw-r--r-- 1 kuangfangjun root 248K Jan  7 19:33 rag_history.wav
-rw-r--r-- 1 kuangfangjun root 173K Jan  7 19:33 rag_math.wav
-rw-r--r-- 1 kuangfangjun root 192K Jan  7 19:33 rag_medical.wav
-rw-r--r-- 1 kuangfangjun root 379K Jan  7 19:33 rag_physics.wav
-rw-r--r-- 1 kuangfangjun root 224K Jan  7 19:33 vietnamese.wav

Hint

If you need the float32 model file, please use:

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-funasr-nano-2025-12-30.tar.bz2

# For Chinese users, you can also use
# wget https://modelscope.cn/models/csukuangfj/asr-models/resolve/master/sherpa-onnx-funasr-nano-2025-12-30.tar.bz2

tar xvf sherpa-onnx-funasr-nano-2025-12-30.tar.bz2
rm sherpa-onnx-funasr-nano-2025-12-30.tar.bz2

ls -lh sherpa-onnx-funasr-nano-2025-12-30

total 3.7G
drwxr-xr-x  5 kuangfangjun root     0 Jan  7 19:27 Qwen3-0.6B
-rw-r--r--  1 kuangfangjun root   253 Jan 13 12:19 README.md
-rw-r--r--  1 kuangfangjun root  594M Jan 13 12:20 embedding.onnx
-rw-r--r--  1 kuangfangjun root  888M Jan 13 12:19 encoder_adaptor.onnx
-rw-r--r--  1 kuangfangjun root  2.3G Jan 13 12:22 llm.fp32.data
-rw-r--r--  1 kuangfangjun root 1011K Jan 13 12:20 llm.fp32.onnx
drwxr-xr-x 27 kuangfangjun root     0 Jan  7 19:27 test_wavs

If you need the float16 model file, please use:

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-funasr-nano-fp16-2025-12-30.tar.bz2

# For Chinese users, you can also use
# wget https://modelscope.cn/models/csukuangfj/asr-models/resolve/master/sherpa-onnx-funasr-nano-fp16-2025-12-30.tar.bz2

tar xvf sherpa-onnx-funasr-nano-fp16-2025-12-30.tar.bz2
rm sherpa-onnx-funasr-nano-fp16-2025-12-30.tar.bz2

ls -lh sherpa-onnx-funasr-nano-fp16-2025-12-30/

total 1.5G
drwxr-xr-x  5 kuangfangjun root    0 Jan  7 19:24 Qwen3-0.6B
-rw-r--r--  1 kuangfangjun root  253 Jan 13 12:26 README.md
-rw-r--r--  1 kuangfangjun root 149M Jan 13 12:26 embedding.int8.onnx
-rw-r--r--  1 kuangfangjun root 227M Jan 13 12:27 encoder_adaptor.int8.onnx
-rw-r--r--  1 kuangfangjun root 1.2G Jan 13 12:27 llm.fp16.onnx
drwxr-xr-x 27 kuangfangjun root    0 Jan  7 19:24 test_wavs

Hint

The test wave files are from

https://funaudiollm.github.io/funasr/

dia_hunan.wav (湖南方言)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_hunan.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_hunan.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_hunan.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.641 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_hunan.wav
{"lang": "", "emotion": "", "event": "", "text": "他总的来讲，孙膑对本怀的理解、文样比庞涓略胜一筹。", "timestamps": [0.00, 0.33, 0.66, 1.00, 1.33, 1.66, 1.99, 2.32, 2.66, 2.99, 3.32, 3.65, 3.98, 4.31, 4.65, 4.98, 5.31, 5.64, 5.97, 6.31, 6.64], "durations": [], "tokens":["他", "总的", "来讲", "，", "孙", "膑", "对", "本", "怀", "的理解", "、", "文", "样", "比", "庞", "涓", "略", "胜", "一", "筹", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.154 s
Real time factor (RTF): 1.154 / 7.012 = 0.165

Wave filename	Content	Ground truth
dia_hunan.wav		但总来讲孙膑对兵法的理解运用比庞涓略胜一筹。

dia_minnan.wav (闽南语)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_minnan.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_minnan.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_minnan.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 2.114 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_minnan.wav
{"lang": "", "emotion": "", "event": "", "text": "嗯，一般有机会吧，因为这个哈开了哈，赚的再厉害，也都挺行啊。", "timestamps": [0.00, 0.37, 0.73, 1.10, 1.46, 1.83, 2.20, 2.56, 2.93, 3.29, 3.66, 4.03, 4.39, 4.76, 5.12, 5.49, 5.85, 6.22, 6.59, 6.95, 7.32, 7.68], "durations": [], "tokens":["嗯", "，", "一般", "有机会", "吧", "，", "因为", "这个", "哈", "开了", "哈", "，", "赚", "的", "再", "厉害", "，", "也都", "挺", "行", "啊", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.329 s
Real time factor (RTF): 1.329 / 8.081 = 0.164

Wave filename	Content	Ground truth
dia_minnan.wav		嗯，下摆若有机会吧，因为即久吼开了吼卷啊遮厉害，会倒贴钱啊。

dia_sh.wav (上海话)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_sh.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_sh.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_sh.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.813 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_sh.wav
{"lang": "", "emotion": "", "event": "", "text": "人跟狗包括人跟动物接触上了才有感情，那么随着社会的富裕。", "timestamps": [0.00, 0.40, 0.81, 1.21, 1.62, 2.02, 2.42, 2.83, 3.23, 3.63, 4.04, 4.44, 4.85, 5.25, 5.65, 6.06, 6.46, 6.87], "durations": [], "tokens":["人", "跟", "狗", "包括", "人", "跟", "动物", "接触", "上了", "才有", "感情", "，", "那么", "随着", "社会", "的", "富裕", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.168 s
Real time factor (RTF): 1.168 / 7.314 = 0.160

Wave filename	Content	Ground truth
dia_sh.wav		人跟狗，包括人跟动物接触长了，全有感情。葛末随了阿拉社会个富裕。

dia_yue.wav (粤语，广东话)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_yue.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_yue.wav

You should see the following output:

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.607 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/dia_yue.wav
{"lang": "", "emotion": "", "event": "", "text": "啲身体好劲啊，跟住咧佢哋有一个人咧就突然可能就有高原反应啦，突然间就啊窒息咗，即系晕晕咗。", "timestamps": [0.00, 0.24, 0.47, 0.71, 0.94, 1.18, 1.41, 1.65, 1.89, 2.12, 2.36, 2.59, 2.83, 3.06, 3.30, 3.54, 3.77, 4.01, 4.24, 4.48, 4.72, 4.95, 5.19, 5.42, 5.66, 5.89, 6.13, 6.37, 6.60, 6.84, 7.07, 7.31, 7.54, 7.78, 8.02, 8.25, 8.49, 8.72, 8.96, 9.19], "durations": [], "tokens":["", "<0xB2>", "身体", "好", "劲", "啊", "，", "跟", "住", "咧", "", "<0xA2>", "", "<0x8B>", "有", "一
个人", "咧", "就", "突然", "可能", "就有", "高原", "反应", "啦", "，", "突然", "间", "就", "啊", "窒息", "", "<0x97>", "，", "即", "系", "晕", "晕", "", "<0x97>", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.749 s
Real time factor (RTF): 1.749 / 9.474 = 0.185

Wave filename	Content	Ground truth
dia_yue.wav		啲身体好劲啊，跟住咧佢哋有一个人咧就突然可能就有高原反应啦，突然间就啊窒息咗，即系晕晕咗。

lyrics.wav (中文歌曲-1)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.640 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics.wav
{"lang": "", "emotion": "", "event": "", "text": "我看到我的身后跟着我的人群，喜欢或恨不一样的神情。我知道这可能就是所谓的成名，我知道必须往前一步一步，不能停。", "timestamps": [0.00, 0.28, 0.57, 0.85, 1.14, 1.42, 1.70, 1.99, 2.27, 2.55, 2.84, 3.12, 3.41, 3.69, 3.97, 4.26, 4.54, 4.82, 5.11, 5.39, 5.68, 5.96, 6.24, 6.53, 6.81, 7.09, 7.38, 7.66, 7.95], "durations": [], "tokens":["我", "看到", "我的", "身后", "跟着", "我的", "人群", "，", "喜欢", "或", "恨", "不一样的", "神情", "。", "我知道", "这", "可能", "就是", "所谓的", "成名", "，", "我知道", "必须", "往前", "一步一步", "，", "不能", "停", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.510 s
Real time factor (RTF): 1.510 / 8.266 = 0.183

Wave filename	Content	Ground truth
lyrics.wav		我看到我的身后盯着我的人群，喜欢或恨不一样的神情，我知道这可能就是所谓的成名，我知道必须往前一步也不能停。

lyrics_2.wav (中文歌曲-2)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_2.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_2.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_2.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.932 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_2.wav
{"lang": "", "emotion": "", "event": "", "text": "明明那么远，为何却感觉离他那么近？闭上眼睛，深深的背出那所有押韵。虽然不听说唱了，但你已学会自信。我代表所有中文说唱歌手向你致敬，如今面对困难的你早已不再抱怨。", "timestamps": [0.00, 0.25, 0.49, 0.74, 0.98, 1.23, 1.47, 1.72, 1.96, 2.21, 2.46, 2.70, 2.95, 3.19, 3.44, 3.68, 3.93, 4.17, 4.42, 4.67, 4.91, 5.16, 5.40, 5.65, 5.89, 6.14, 6.38, 6.63, 6.88, 7.12, 7.37, 7.61, 7.86, 8.10, 8.35, 8.59, 8.84, 9.08, 9.33, 9.58, 9.82, 10.07, 10.31, 10.56, 10.80, 11.05, 11.29, 11.54, 11.79, 12.03, 12.28, 12.52, 12.77, 13.01, 13.26, 13.50], "durations": [], "tokens":["明明", "那么", "远", "，", "为何", "却", "感觉", "离", "他", "那么", "近", "？", "闭", "上", "眼睛", "，", "深深的", "背", "出", "那", "所有", "押", "韵", "。", "虽然", "不", "听说", "唱", "了", "，", "但", "你", "已", "学会", "自信", "。", "我", "代表", "所有", "中文", "说", "唱", "歌手", "向", "你", "致敬", "，", "如今", "面对", "困难", "的", "你", "早已", "不再", "抱怨", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 2.627 s
Real time factor (RTF): 2.627 / 13.769 = 0.191

Wave filename	Content	Ground truth
lyrics_2.wav		明明那么远，为何却感觉离他那么近？闭上眼，你甚至能背出他所有押韵。虽然不听说唱了，但你已学会自信。我代表所有中文说唱歌手向你致敬。如今面对困难的你，早已不再抱怨。

lyrics_3.wav (中文歌曲-3)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_3.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_3.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_3.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.704 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_3.wav
{"lang": "", "emotion": "", "event": "", "text": "你听啊，秋末的落叶，你听它叹息着离别，只剩我独自领略海与山风和月。你听啊。", "timestamps": [0.00, 0.54, 1.09, 1.63, 2.18, 2.72, 3.26, 3.81, 4.35, 4.90, 5.44, 5.98, 6.53, 7.07, 7.62, 8.16, 8.70, 9.25, 9.79, 10.34, 10.88, 11.43, 11.97, 12.51, 13.06, 13.60, 14.15, 14.69, 15.23, 15.78, 16.32, 16.87], "durations": [], "tokens":["你", "听", "啊", "，", "秋", "末", "的", "落叶", "，", "你", "听", "它", "叹息", "着", "离", "别", "，", "只剩", "我", "独自", "领略", "海", "与", "山", "风", "和", "月", "。", "你", "听", "啊", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 2.744 s
Real time factor (RTF): 2.744 / 17.461 = 0.157

Wave filename	Content	Ground truth
lyrics_3.wav		你听啊秋末的落叶，你听它叹息着离别，只剩我独自领略海与山风和月，你听啊。

lyrics_en_1.wav (英文歌曲-1)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_1.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_1.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_1.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.989 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_1.wav
{"lang": "", "emotion": "", "event": "", "text": "When I was young, I'd listen to the radio waiting for my favorite songs. When they played, I'd sing along. It made me smile.", "timestamps": [0.00, 0.75, 1.50, 2.25, 3.01, 3.76, 4.51, 5.26, 6.01, 6.76, 7.51, 8.26, 9.02, 9.77, 10.52, 11.27, 12.02, 12.77, 13.52, 14.27, 15.03, 15.78, 16.53, 17.28, 18.03, 18.78, 19.53, 20.28, 21.04, 21.79, 22.54], "durations": [], "tokens":["When", "I", "was", "young", ",", "I", "'d", "listen", "to", "the", "radio", "waiting", "for", "my", "favorite", "songs", ".", "When", "they", "played", ",", "I", "'d", "sing", "along", ".", "It", "made", "me", "smile", "."], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 3.729 s
Real time factor (RTF): 3.729 / 23.341 = 0.160

Wave filename	Content	Ground truth
lyrics_en_1.wav		When I was young I'd listen to the radio. Waiting for my favorite songs. When they played I'd sing along. It made me smile.

lyrics_en_2.wav (英文歌曲-2)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_2.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_2.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_2.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.707 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_2.wav
{"lang": "", "emotion": "", "event": "", "text": "I see your monsters, I see your pain, tell me your problems, I'll chase them away. I'll be your lighthouse, I'll make it okay. When I see your monsters, I'll stand there so brave and chase them all away.", "timestamps": [0.00, 0.42, 0.83, 1.25, 1.67, 2.08, 2.50, 2.92, 3.33, 3.75, 4.17, 4.58, 5.00, 5.42, 5.83, 6.25, 6.67, 7.08, 7.50, 7.92, 8.33, 8.75, 9.17, 9.58, 10.00, 10.42, 10.84, 11.25, 11.67, 12.09, 12.50, 12.92, 13.34, 13.75, 14.17, 14.59, 15.00, 15.42, 15.84, 16.25, 16.67, 17.09, 17.50, 17.92, 18.34, 18.75, 19.17, 19.59, 20.00, 20.42, 20.84, 21.25], "durations": [], "tokens":["I", "see", "your", "monsters", ",", "I", "see", "your", "pain", ",", "tell", "me", "your", "problems", ",", "I", "'ll", "chase", "them", "away", ".", "I", "'ll", "be", "your", "l", "ighthouse", ",", "I", "'ll", "make", "it", "okay", ".", "When", "I", "see", "your", "monsters", ",", "I", "'ll", "stand", "there", "so", "brave", "and", "chase", "them", "all", "away", "."], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 3.741 s
Real time factor (RTF): 3.741 / 21.711 = 0.172

Wave filename	Content	Ground truth
lyrics_en_2.wav		I see your monsters. I see your pain. Tell me your problems; I'll chase them away. I'll be your lighthouse. I'll make it okay. When I see your monsters, I'll stand there so brave and chase them all away.

lyrics_en_3.wav (英文歌曲-3)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_3.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_3.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_3.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.758 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/lyrics_en_3.wav
{"lang": "", "emotion": "", "event": "", "text": "An empty street, an empty house, a hole inside my heart. I'm all alone, the rooms are getting smaller. I wonder how, I wonder why, I wonder where they are. The days we had, the songs we sang together.", "timestamps": [0.00, 0.50, 0.99, 1.49, 1.99, 2.49, 2.98, 3.48, 3.98, 4.47, 4.97, 5.47, 5.96, 6.46, 6.96, 7.45, 7.95, 8.45, 8.95, 9.44, 9.94, 10.44, 10.93, 11.43, 11.93, 12.43, 12.92, 13.42, 13.92, 14.41, 14.91, 15.41, 15.90, 16.40, 16.90, 17.40, 17.89, 18.39, 18.89, 19.38, 19.88, 20.38, 20.87, 21.37, 21.87, 22.36, 22.86, 23.36, 23.86, 24.35], "durations": [], "tokens":["An", "empty", "street", ",", "an", "empty", "house", ",", "a", "hole", "inside", "my", "heart", ".", "I", "'m", "all", "alone", ",", "the", "rooms", "are", "getting", "smaller", ".", "I", "wonder", "how", ",", "I", "wonder", "why", ",", "I", "wonder", "where", "they", "are", ".", "The", "days", "we", "had", ",", "the", "songs", "we", "sang", "together", "."], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 4.418 s
Real time factor (RTF): 4.418 / 24.882 = 0.178

Wave filename	Content	Ground truth
lyrics_en_3.wav		An empty street, an empty house, a hole inside my heart. I'm all alone and the rooms are getting smaller. I wonder how, I wonder why, I wonder where they are. The days we had, the songs we sang together.

noise_en.wav (英文歌曲-3)

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/noise_en.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/noise_en.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/noise_en.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.661 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/noise_en.wav
{"lang": "", "emotion": "", "event": "", "text": "So what's interesting here is, I feel that you know brands knowing this when people sort of speak to the voice assistants at home, and if you want to be the brand.", "timestamps": [0.00, 0.29, 0.59, 0.88, 1.17, 1.47, 1.76, 2.06, 2.35, 2.64, 2.94, 3.23, 3.52, 3.82, 4.11, 4.40, 4.70, 4.99, 5.28, 5.58, 5.87, 6.17, 6.46, 6.75, 7.05, 7.34, 7.63, 7.93, 8.22, 8.51, 8.81, 9.10, 9.40, 9.69, 9.98, 10.28], "durations": [], "tokens":["So", "what", "'s", "interesting", "here", "is", ",", "I", "feel", "that", "you", "know", "brands", "knowing", "this", "when", "people", "sort", "of", "speak", "to", "the", "voice", "assistants", "at", "home", ",", "and", "if", "you", "want", "to", "be", "the", "brand", "."], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.968 s
Real time factor (RTF): 1.968 / 10.568 = 0.186

Wave filename	Content	Ground truth
noise_en.wav		So what's interesting here is I feel that you know brands knowing this when people sort of speak to the voice assistance at home and if you want to be the brand.

far_2.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_2.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_2.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_2.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.844 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_2.wav
{"lang": "", "emotion": "", "event": "", "text": "然后被灌顶了渣南县的城防，跑了半截的。那么前方即将到达省公路站，左边是八号线。", "timestamps": [0.00, 0.23, 0.46, 0.68, 0.91, 1.14, 1.37, 1.60, 1.83, 2.05, 2.28, 2.51, 2.74, 2.97, 3.20, 3.42, 3.65, 3.88, 4.11, 4.34, 4.57, 4.79, 5.02, 5.25, 5.48, 5.71, 5.94, 6.16, 6.39, 6.62], "durations": [], "tokens":["然后", "被", "灌", "顶", "了", "渣", "南", "县", "的", "城", "防", "，", "跑了", "半", "截", "的", "。", "那么", "前方", "即将", "到达", "省", "公路", "站", "，", "左边", "是", "八", "号线", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.296 s
Real time factor (RTF): 1.296 / 6.850 = 0.189

Wave filename	Content	Ground truth
far_2.wav		然后被冠以了渣男线的称号，好了，不管这个，那么前方即将到达沈杜公路站，左边是8号线。

far_3.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_3.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_3.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_3.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.673 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_3.wav
{"lang": "", "emotion": "", "event": "", "text": "周末要不要去露营？最近天气超舒服，露营我怕虫子咬，而且晚上睡帐篷会不会很冷啊？放心，我借了专业装备，还有暖宝宝，再带点火锅食材，边吃边看星星超惬意。", "timestamps": [0.00, 0.40, 0.81, 1.21, 1.61, 2.02, 2.42, 2.82, 3.23, 3.63, 4.04, 4.44, 4.84, 5.25, 5.65, 6.05, 6.46, 6.86, 7.26, 7.67, 8.07, 8.47, 8.88, 9.28, 9.68, 10.09, 10.49, 10.90, 11.30, 11.70, 12.11, 12.51, 12.91, 13.32, 13.72, 14.12, 14.53, 14.93, 15.33, 15.74, 16.14, 16.54, 16.95, 17.35, 17.75, 18.16, 18.56, 18.97, 19.37, 19.77, 20.18, 20.58, 20.98, 21.39], "durations": [], "tokens":["周末", "要不要", "去", "露", "营", "？", "最近", "天气", "超", "舒服", "，", "露", "营", "我", "怕", "虫", "子", "咬", "，", "而且", "晚上", "睡", "帐篷", "会不会", "很", "冷", "啊", "？", "放心", "，", "我", "借", "了", "专业", "装备", "，", "还有", "暖", "宝宝", "，", "再", "带", "点", "火锅", "食材", "，", "边", "吃", "边", "看", "星星", "超", "惬意", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 3.809 s
Real time factor (RTF): 3.809 / 21.804 = 0.175

Wave filename	Content	Ground truth
far_3.wav		周末要不要去露营，最近天气超舒服，露营？我怕虫子咬，而且晚上睡帐篷会不会很冷啊？放心，我借了专业装备还有暖宝宝，再带点火锅食材，边吃边看星星超惬意。

far_4.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_4.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_4.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_4.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.859 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_4.wav
{"lang": "", "emotion": "", "event": "", "text": "唯一的遗憾就是他那个八宝鸭还有烤鸭子没吃上，估计得提前预定吧，只能怪我自己没有做好功课。", "timestamps": [0.00, 0.31, 0.63, 0.94, 1.25, 1.56, 1.88, 2.19, 2.50, 2.81, 3.13, 3.44, 3.75, 4.07, 4.38, 4.69, 5.00, 5.32, 5.63, 5.94, 6.26, 6.57, 6.88, 7.19, 7.51, 7.82, 8.13, 8.44, 8.76], "durations": [], "tokens":["唯一的", "遗憾", "就是", "他", "那个", "八", "宝", "鸭", "还有", "烤", "鸭", "子", "没", "吃", "上", "，", "估计", "得", "提前", "预定", "吧", "，", "只能", "怪", "我自己", "没有", "做好", "功课", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.570 s
Real time factor (RTF): 1.570 / 9.079 = 0.173

Wave filename	Content	Ground truth
far_4.wav		唯一的遗憾就是他那个八宝鸭还有烤鸭都没吃上, 估计得提前预定吧, 只能怪我自己没有做好功课.

far_5.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_5.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_5.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_5.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.714 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/far_5.wav
{"lang": "", "emotion": "", "event": "", "text": "别紧张，我只是我是在这边逛街，然后看到你们在这边拍照，想跟你交个朋友，认识一下。", "timestamps": [0.00, 0.36, 0.71, 1.07, 1.42, 1.78, 2.13, 2.49, 2.84, 3.20, 3.56, 3.91, 4.27, 4.62, 4.98, 5.33, 5.69, 6.05, 6.40, 6.76, 7.11, 7.47, 7.82, 8.18, 8.53], "durations": [], "tokens":["别", "紧张", "，", "我只是", "我", "是在", "这边", "逛街", "，", "然后", "看到", "你们", "在这", "边", "拍照", "，", "想", "跟你", "交", "个", "朋友", "，", "认识", "一下", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.503 s
Real time factor (RTF): 1.503 / 8.917 = 0.169

Wave filename	Content	Ground truth
far_5.wav		别紧张, 我只是我是在这边逛街, 然后看到你们在这边拍照, 想跟你交个朋友, 认识一下.

rag_chemistry.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_chemistry.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_chemistry.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_chemistry.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.899 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_chemistry.wav
{"lang": "", "emotion": "", "event": "", "text": "比如说只在当时被认为是一种含氧酸盐。", "timestamps": [0.00, 0.62, 1.23, 1.85, 2.47, 3.09, 3.70, 4.32, 4.94, 5.56, 6.17], "durations": [], "tokens":["比如说", "只", "在", "当时", "被认为", "是一种", "含", "氧", "酸", "盐", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.081 s
Real time factor (RTF): 1.081 / 6.840 = 0.158

Wave filename	Content	Ground truth
rag_chemistry.wav		比如说酯在当时被认为是一种含氧酸盐

rag_history.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_history.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_history.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_history.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.776 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_history.wav
{"lang": "", "emotion": "", "event": "", "text": "由罗马皇帝钦点的犹地亚王大希律王统治期间", "timestamps": [0.00, 0.49, 0.98, 1.48, 1.97, 2.46, 2.95, 3.44, 3.93, 4.43, 4.92, 5.41, 5.90, 6.39, 6.89, 7.38], "durations": [], "tokens":["由", "罗马", "皇帝", "钦", "点", "的", "犹", "地", "亚", "王", "大", "希", "律", "王", "统治", "期间"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.261 s
Real time factor (RTF): 1.261 / 7.920 = 0.159

Wave filename	Content	Ground truth
rag_history.wav		由罗马皇帝钦点的犹地亚王大希律王统治期间

rag_math.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_math.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_math.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_math.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.943 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_math.wav
{"lang": "", "emotion": "", "event": "", "text": "对微分形式的积分是微分几何中的基本概念。", "timestamps": [0.00, 0.39, 0.78, 1.17, 1.56, 1.95, 2.34, 2.73, 3.13, 3.52, 3.91, 4.30, 4.69, 5.08], "durations": [], "tokens":["对", "微", "分", "形式", "的", "积分", "是", "微", "分", "几何", "中的", "基本", "概念", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 0.946 s
Real time factor (RTF): 0.946 / 5.520 = 0.171

Wave filename	Content	Ground truth
rag_math.wav		对微分形式的积分是微分几何中的基本概念

rag_medical.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_medical.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_medical.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_medical.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.744 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_medical.wav
{"lang": "", "emotion": "", "event": "", "text": "肾脏中肾小球囊上的细胞膜孔隙很小。", "timestamps": [0.00, 0.47, 0.93, 1.40, 1.87, 2.33, 2.80, 3.27, 3.74, 4.20, 4.67, 5.14, 5.60], "durations": [], "tokens":["肾脏", "中", "肾", "小", "球", "囊", "上的", "细胞", "膜", "孔", "隙", "很小", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.006 s
Real time factor (RTF): 1.006 / 6.120 = 0.164

Wave filename	Content	Ground truth
rag_medical.wav		肾脏中肾小球囊上的细胞膜孔隙很小

rag_physics.wav

To decode the test file ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_physics.wav:

./build/bin/sherpa-onnx-offline \
  --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx \
  --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx \
  --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B \
  --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx \
  ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_physics.wav

You should see the following output:

/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:373 ./build/bin/sherpa-onnx-offline --funasr-nano-encoder-adaptor=./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx --funasr-nano-llm=./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx --funasr-nano-tokenizer=./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B --funasr-nano-embedding=./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx ./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_physics.wav 

OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0, normalize_samples=True, snip_edges=False), model_config=OfflineModelConfig(transducer=OfflineTransducerModelConfig(encoder_filename="", decoder_filename="", joiner_filename=""), paraformer=OfflineParaformerModelConfig(model=""), nemo_ctc=OfflineNemoEncDecCtcModelConfig(model=""), whisper=OfflineWhisperModelConfig(encoder="", decoder="", language="", task="transcribe", tail_paddings=-1), fire_red_asr=OfflineFireRedAsrModelConfig(encoder="", decoder=""), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), sense_voice=OfflineSenseVoiceModelConfig(model="", language="auto", use_itn=False), moonshine=OfflineMoonshineModelConfig(preprocessor="", encoder="", uncached_decoder="", cached_decoder=""), dolphin=OfflineDolphinModelConfig(model=""), canary=OfflineCanaryModelConfig(encoder="", decoder="", src_lang="", tgt_lang="", use_pnc=True), omnilingual=OfflineOmnilingualAsrCtcModelConfig(model=""), funasr_nano=OfflineFunASRNanoModelConfig(encoder_adaptor="./sherpa-onnx-funasr-nano-int8-2025-12-30/encoder_adaptor.int8.onnx", llm="./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx", embedding="./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx", tokenizer="./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B", system_prompt="You are a helpful assistant.", user_prompt="语音转写：", max_new_tokens=512, temperature=1e-06, top_p=0.8, seed=42), medasr=OfflineMedAsrCtcModelConfig(model=""), telespeech_ctc="", tokens="", num_threads=2, debug=False, provider="cpu", model_type="", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5, lodr_scale=0.01, lodr_fst="", lodr_backoff_id=-1), ctc_fst_decoder_config=OfflineCtcFstDecoderConfig(graph="", max_active=3000), decoding_method="greedy_search", max_active_paths=4, hotwords_file="", hotwords_score=1.5, blank_penalty=0, rule_fsts="", rule_fars="", hr=HomophoneReplacerConfig(lexicon="", rule_fsts=""))
Creating recognizer ...
recognizer created in 1.794 s
Started
Done!

./sherpa-onnx-funasr-nano-int8-2025-12-30/test_wavs/rag_physics.wav
{"lang": "", "emotion": "", "event": "", "text": "根据碰撞理论，月面样本缺少挥发性物质。", "timestamps": [0.00, 1.01, 2.01, 3.02, 4.02, 5.03, 6.03, 7.04, 8.05, 9.05, 10.06, 11.06], "durations": [], "tokens":["根据", "碰撞", "理论", "，", "月", "面", "样本", "缺少", "挥发", "性", "物质", "。"], "ys_log_probs": [], "words": []}
----
num threads: 2
decoding method: greedy_search
Elapsed seconds: 1.743 s
Real time factor (RTF): 1.743 / 12.120 = 0.144

Wave filename	Content	Ground truth
rag_physics.wav		根据碰撞理论月面样本缺少挥发性物质