Models
sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04 (支持非常多种方言)
Hint
这个模型支持很多种方言。
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-telespeech-ctc-int8-zh-2024-06-04.tar.bz2
tar xvf sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04.tar.bz2
rm sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04.tar.bz2
Please check that the file sizes of the pre-trained models are correct. See
the file sizes of *.onnx files below.
$ ls -lh *.onnx
-rw-r--r-- 1 fangjun staff 325M Jun 4 11:56 model.int8.onnx
Decode wave files
Hint
It supports decoding only wave files of a single channel with 16-bit encoded samples, while the sampling rate does not need to be 16 kHz.
cd /path/to/sherpa-onnx
./build/bin/sherpa-onnx-offline \
--telespeech-ctc=./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/model.int8.onnx \
--tokens=./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/tokens.txt \
--model-type=telespeech_ctc \
--num-threads=1 \
./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/3-sichuan.wav \
./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/4-tianjin.wav \
./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/5-henan.wav
Note
Please use ./build/bin/Release/sherpa-onnx-offline.exe for Windows.
Caution
If you use Windows and get encoding issues, please run:
CHCP 65001
in your commandline.
You should see the following output:
/Users/fangjun/open-source/sherpa-onnx/sherpa-onnx/csrc/parse-options.cc:Read:361 ./build/bin/sherpa-onnx-offline --telespeech-ctc=./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/model.int8.onnx --tokens=./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/tokens.txt --model-type=telespeech_ctc --num-threads=1 ./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/3-sichuan.wav ./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/4-tianjin.wav ./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/5-henan.wav
OfflineRecognizerConfig(feat_config=FeatureExtractorConfig(sampling_rate=16000, feature_dim=80, low_freq=20, high_freq=-400, dither=0), 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), tdnn=OfflineTdnnModelConfig(model=""), zipformer_ctc=OfflineZipformerCtcModelConfig(model=""), wenet_ctc=OfflineWenetCtcModelConfig(model=""), telespeech_ctc="./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/model.int8.onnx", tokens="./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/tokens.txt", num_threads=1, debug=False, provider="cpu", model_type="telespeech_ctc", modeling_unit="cjkchar", bpe_vocab=""), lm_config=OfflineLMConfig(model="", scale=0.5), 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)
Creating recognizer ...
Started
Done!
./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/3-sichuan.wav
{"text": "自己就是在那个在那个就是在情界里面就是感觉演得特别好就是好像很真实样知道吧", "timestamps": [0.08, 0.36, 0.52, 0.72, 0.92, 1.16, 1.36, 1.88, 2.20, 2.36, 3.16, 3.28, 3.40, 3.60, 3.80, 3.92, 4.08, 4.24, 4.40, 4.56, 4.76, 5.16, 5.32, 5.44, 5.64, 5.76, 5.88, 6.04, 6.16, 6.28, 6.40, 6.60, 6.88, 7.12, 7.40, 7.52, 7.64], "tokens":["自", "己", "就", "是", "在", "那", "个", "在", "那", "个", "就", "是", "在", "情", "界", "里", "面", "就", "是", "感", "觉", "演", "得", "特", "别", "好", "就", "是", "好", "像", "很", "真", "实", "样", "知", "道", "吧"]}
----
./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/4-tianjin.wav
{"text": "他就每个人手法就这意思法律意识太单薄了而且就是也不顾及到别人的感受", "timestamps": [0.36, 0.56, 1.04, 1.16, 1.24, 1.64, 1.88, 2.24, 2.40, 2.60, 2.80, 3.12, 3.32, 3.64, 3.80, 3.96, 4.16, 4.44, 4.68, 4.80, 5.00, 5.16, 5.28, 6.12, 6.28, 6.44, 6.60, 6.72, 6.88, 7.04, 7.12, 7.32, 7.52], "tokens":["他", "就", "每", "个", "人", "手", "法", "就", "这", "意", "思", "法", "律", "意", "识", "太", "单", "薄", "了", "而", "且", "就", "是", "也", "不", "顾", "及", "到", "别", "人", "的", "感", "受"]}
----
./sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04/test_wavs/5-henan.wav
{"text": "他这个管一都通到有时候都通到七八层楼高然它这个管箱就可交到那个那珠子上", "timestamps": [0.04, 0.12, 0.24, 0.40, 1.00, 1.24, 1.44, 1.68, 2.32, 2.48, 2.60, 2.64, 2.80, 3.00, 3.16, 3.32, 3.52, 3.68, 3.92, 5.00, 5.16, 5.28, 5.32, 5.44, 5.84, 6.00, 6.12, 6.48, 6.68, 6.84, 7.00, 7.16, 7.32, 7.56, 7.68], "tokens":["他", "这", "个", "管", "一", "都", "通", "到", "有", "时", "候", "都", "通", "到", "七", "八", "层", "楼", "高", "然", "它", "这", "个", "管", "箱", "就", "可", "交", "到", "那", "个", "那", "珠", "子", "上"]}
----
num threads: 1
decoding method: greedy_search
Elapsed seconds: 3.406 s
Real time factor (RTF): 3.406 / 23.634 = 0.144
Note
The feature_dim=80 is incorrect in the above logs. The actual value is 40.
Hint
There is also a float32 model. Please see https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-telespeech-ctc-zh-2024-06-04.tar.bz2