Streaming (Online) ASR Models

Table of Contents

• Transducer (Zipformer)
• Nemotron (NeMo Transducer)
• Streaming Paraformer
• Zipformer2 CTC
• T-One CTC

sherpa-onnx supports several streaming ASR model families through the SherpaOnnxCreateOnlineRecognizer() API. Configure exactly one model family by filling in the corresponding sub-struct of SherpaOnnxOnlineModelConfig.

See also

SherpaOnnxCreateOnlineRecognizer, SherpaOnnxOnlineRecognizerConfig, SherpaOnnxOnlineModelConfig

Transducer (Zipformer)

Streaming Zipformer transducer models are the most widely used streaming ASR models in sherpa-onnx.

SherpaOnnxOnlineRecognizerConfig config;
memset(&config, 0, sizeof(config));
config.feat_config.sample_rate = 16000;
config.feat_config.feature_dim = 80;
config.model_config.transducer.encoder =
    "./sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/"
    "encoder-epoch-99-avg-1.int8.onnx";
config.model_config.transducer.decoder =
    "./sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/"
    "decoder-epoch-99-avg-1.onnx";
config.model_config.transducer.joiner =
    "./sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/"
    "joiner-epoch-99-avg-1.int8.onnx";
config.model_config.tokens =
    "./sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/tokens.txt";
config.model_config.provider = "cpu";
config.model_config.num_threads = 1;
config.decoding_method = "greedy_search";
 
const SherpaOnnxOnlineRecognizer *recognizer =
    SherpaOnnxCreateOnlineRecognizer(&config);

Model package: sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20

Example source: streaming-zipformer-c-api.c

Nemotron (NeMo Transducer)

Nemotron is an NVIDIA NeMo streaming transducer model. The runtime auto-detects NeMo transducer models from the decoder's output node count, so no model_type string is needed.

config.model_config.transducer.encoder =
    "./sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25/"
    "encoder.int8.onnx";
config.model_config.transducer.decoder =
    "./sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25/"
    "decoder.int8.onnx";
config.model_config.transducer.joiner =
    "./sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25/"
    "joiner.int8.onnx";
config.model_config.tokens =
    "./sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25/"
    "tokens.txt";
config.model_config.provider = "cpu";
config.model_config.num_threads = 1;
config.decoding_method = "greedy_search";

Model package: sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25

Available variants: 80ms, 160ms, 560ms, 1120ms chunk sizes.

Example source: streaming-nemotron-c-api.c

Streaming Paraformer

Streaming Paraformer is a non-autoregressive streaming ASR model from FunASR.

config.model_config.paraformer.encoder =
    "./sherpa-onnx-streaming-paraformer-bilingual-zh-en/encoder.int8.onnx";
config.model_config.paraformer.decoder =
    "./sherpa-onnx-streaming-paraformer-bilingual-zh-en/decoder.int8.onnx";
config.model_config.tokens =
    "./sherpa-onnx-streaming-paraformer-bilingual-zh-en/tokens.txt";
config.model_config.provider = "cpu";
config.model_config.num_threads = 1;
config.decoding_method = "greedy_search";

Model package: sherpa-onnx-streaming-paraformer-bilingual-zh-en

Example source: streaming-paraformer-c-api.c

Zipformer2 CTC

Streaming Zipformer2 CTC models use CTC decoding instead of transducer decoding.

config.model_config.zipformer2_ctc.model =
    "./sherpa-onnx-streaming-zipformer-ctc-multi-zh-hans-2023-12-13/"
    "ctc-epoch-20-avg-1-chunk-16-left-128.int8.onnx";
config.model_config.tokens =
    "./sherpa-onnx-streaming-zipformer-ctc-multi-zh-hans-2023-12-13/"
    "tokens.txt";
config.model_config.provider = "cpu";
config.model_config.num_threads = 1;
config.decoding_method = "greedy_search";

Model package: sherpa-onnx-streaming-zipformer-ctc-multi-zh-hans-2023-12-13

Example source: streaming-ctc-buffered-tokens-c-api.c

T-One CTC

T-One is a streaming CTC model for Russian.

config.model_config.t_one_ctc.model =
    "./sherpa-onnx-streaming-t-one-russian-2025-09-08/model.onnx";
config.model_config.tokens =
    "./sherpa-onnx-streaming-t-one-russian-2025-09-08/tokens.txt";
config.model_config.provider = "cpu";
config.model_config.num_threads = 1;
config.decoding_method = "greedy_search";

Model package: sherpa-onnx-streaming-t-one-russian-2025-09-08

Example source: streaming-t-one-ctc-c-api.c