update

游雁

2024-07-17 a836eca98e30fa67d45167dac40f359ae42d42ec

 README.md

@@ -93,7 +93,7 @@

|                                                                                                         Model Name                                                                                                         |                                   Task Details                                   |          Training Data           | Parameters |
|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------------:|:--------------------------------:|:----------:|
|                                       SenseVoiceSmall <br> ([⭐](https://www.modelscope.cn/models/iic/SenseVoiceSmall)  [🤗](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) )                                          | multiple speech understanding capabilities, including ASR, LID, SER, and AED.    |           400000 hours           |   330M     |
|                                       SenseVoiceSmall <br> ([⭐](https://www.modelscope.cn/models/iic/SenseVoiceSmall)  [🤗](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) )                                          | multiple speech understanding capabilities, including ASR, ITN, LID, SER, and AED, support languages such as zh, yue, en, ja, ko   |           300000 hours           |   234M     |
|          paraformer-zh <br> ([⭐](https://www.modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary)  [🤗](https://huggingface.co/funasr/paraformer-zh) )           |                speech recognition, with timestamps, non-streaming                |      60000 hours, Mandarin       |    220M    |
| <nobr>paraformer-zh-streaming <br> ( [⭐](https://modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/summary) [🤗](https://huggingface.co/funasr/paraformer-zh-streaming) )</nobr> |                          speech recognition, streaming                           |      60000 hours, Mandarin       |    220M    |
|               paraformer-en <br> ( [⭐](https://www.modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-en-16k-common-vocab10020/summary) [🤗](https://huggingface.co/funasr/paraformer-en) )                |              speech recognition, without timestamps, non-streaming               |       50000 hours, English       |    220M    |
@@ -129,6 +129,44 @@
Notes: Support recognition of single audio file, as well as file list in Kaldi-style wav.scp format: `wav_id wav_pat`

### Speech Recognition (Non-streaming)
#### SenseVoice
```python
from funasr import AutoModel
from funasr.utils.postprocess_utils import rich_transcription_postprocess

model_dir = "iic/SenseVoiceSmall"

model = AutoModel(
    model=model_dir,
    vad_model="fsmn-vad",
    vad_kwargs={"max_single_segment_time": 30000},
    device="cuda:0",
)

# en
res = model.generate(
    input=f"{model.model_path}/example/en.mp3",
    cache={},
    language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"
    use_itn=True,
    batch_size_s=60,
    merge_vad=True,  #
    merge_length_s=15,
)
text = rich_transcription_postprocess(res[0]["text"])
print(text)
```
Parameter Descriptions:
- `model_dir`: The name of the model, or the model's path on the local disk.
- `trust_remote_code`:
  - When set to `True`, this indicates that the model's code implementation should be loaded from the location specified by `remote_code`, which points to the exact code for the model (for example, `model.py` in the current directory). It supports absolute paths, relative paths, and network URLs.
  - When set to `False`, this signifies that the model's code implementation is the integrated version within [FunASR](https://github.com/modelscope/FunASR). In this case, any modifications to `model.py` in the current directory will not take effect because the version loaded is the internal one from FunASR. For the model code, [click here to view](https://github.com/modelscope/FunASR/tree/main/funasr/models/sense_voice).
- `max_single_segment_time`: The maximum length of audio segments that the `vad_model` can cut, measured in milliseconds (ms).
- `use_itn`: Indicates whether the output should include punctuation and inverse text normalization.
- `batch_size_s`: Represents a dynamic batch size where the total duration of the audio in the batch is measured in seconds (s).
- `merge_vad`: Whether to concatenate short audio fragments cut by the vad model, with the merged length being `merge_length_s`, measured in seconds (s).

#### Paraformer
```python
from funasr import AutoModel
# paraformer-zh is a multi-functional asr model