FunASR/model_zoo/huggingface_models.md

Pretrained Models on Huggingface

Model License

The FunASR toolkit uses the MIT software license. Model weights have separate terms: check the exact checkpoint's model card, license file and revision before use or redistribution. The FunASR model agreement applies only to models whose published terms adopt it, not every entry in this catalogue. Third-party models, including OpenMOSS MOSS-Transcribe-Diarize, retain their original authorship and model-specific licenses.

For model selection and deployment boundaries, start with the Model Zoo guide and MOSS integration.

Model Zoo

Here we provided several pretrained models on different datasets. The details of models and datasets can be found on ModelScope.

Speech Recognition Models

Paraformer Models

Model Name	Language	Training Data	Vocab Size	Parameter	Offline/Online	Notes
Paraformer-large	CN & EN	Alibaba Speech Data (60000hours)	8404	220M	Offline	Duration of input wav <= 20s

UniASR Models

Conformer Models

RNN-T Models

Multi-talker Speech Recognition Models

MFCCA Models

Voice Activity Detection Models

Model Name	Training Data	Parameters	Sampling Rate	Notes
FSMN-VAD	Alibaba Speech Data (5000hours)	0.4M	16000

Punctuation Restoration Models

Model Name	Training Data	Parameters	Vocab Size	Offline/Online	Notes
CT-Transformer	Alibaba Text Data	70M	272727	Offline	offline punctuation model

Language Models

Speaker Verification Models

Speaker diarization Models

Third-party Unified Transcription and Diarization

Model Name	Publisher	Parameters	Notes
MOSS-Transcribe-Diarize	OpenMOSS	See official model card	Third-party Apache-2.0 model. One offline request returns transcription, timestamps, and speaker labels; the FunASR integration does not require an external VAD service. See the deployment guide.

Timestamp Prediction Models