model-whisper-large-v3/README.md

---
tasks:
- auto-speech-recognition
domain:
- audio
model-type:
- autoregressive
frameworks:
- pytorch
backbone:
- transformer/conformer
metrics:
- CER
license: Apache License 2.0
language: 
- multilingual
tags:
- FunASR
- Whisper
datasets:
  train:
  - 680,000 hour
  test:
  - test
indexing:
   results:
   - task:
       name: Automatic Speech Recognition
     dataset:
       name: 680,000 hour
     metrics:
       - type: CER
         value: 8.53%  # float
         description: greedy search, withou lm, avg.
         args: default
       - type: RTF
         value: 0.0251  # float
         description: GPU inference on V100
         args: batch_size=1
widgets:
  - task: auto-speech-recognition
    model_revision: v1.0.0
    inputs:
      - type: audio
        name: input
        title: 音频
    examples:
      - name: 1
        title: 示例1
        inputs:
          - name: input
            data: git://example/asr_example.wav
    inferencespec:
      cpu: 8 #CPU数量
      memory: 4096
---


# Whisper模型介绍

## <strong>[ModelScope-FunASR](https://github.com/alibaba-damo-academy/FunASR)</strong>
<strong>[FunASR](https://github.com/alibaba-damo-academy/FunASR)</strong>希望在语音识别方面建立学术研究和工业应用之间的桥梁。通过支持在ModelScope上发布的工业级语音识别模型的训练和微调，研究人员和开发人员可以更方便地进行语音识别模型的研究和生产，并促进语音识别生态系统的发展。

[**最新动态**](https://github.com/alibaba-damo-academy/FunASR#whats-new) 
| [**环境安装**](https://github.com/alibaba-damo-academy/FunASR#installation)
| [**介绍文档**](https://alibaba-damo-academy.github.io/FunASR/en/index.html)
| [**中文教程**](https://github.com/alibaba-damo-academy/FunASR/wiki#funasr%E7%94%A8%E6%88%B7%E6%89%8B%E5%86%8C)
| [**服务部署**](https://github.com/alibaba-damo-academy/FunASR/tree/main/funasr/runtime)
| [**模型库**](https://github.com/alibaba-damo-academy/FunASR/blob/main/docs/model_zoo/modelscope_models.md)
| [**联系我们**](https://github.com/alibaba-damo-academy/FunASR#contact)


## 基于ModelScope进行推理

- 推理支持音频格式如下：
  - wav文件路径，例如：data/test/audios/asr_example.wav
  - pcm文件路径，例如：data/test/audios/asr_example.pcm
  - wav文件url，例如：https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav
  - wav二进制数据，格式bytes，例如：用户直接从文件里读出bytes数据或者是麦克风录出bytes数据。
  - 已解析的audio音频，例如：audio, rate = soundfile.read("asr_example_zh.wav")，类型为numpy.ndarray或者torch.Tensor。
  - wav.scp文件，需符合如下要求：

```sh
cat wav.scp
asr_example1  data/test/audios/asr_example1.wav
asr_example2  data/test/audios/asr_example2.wav
...
```

- 若输入格式wav文件url，api调用方式可参考如下范例：

```python
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks

inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='iic/Whisper-large-v3')

rec_result = inference_pipeline(input='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav', language=None)
print(rec_result)
```

- 输入音频为pcm格式，调用api时需要传入音频采样率参数fs，例如：

```python
rec_result = inference_pipeline(input='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.pcm', fs=16000)
```

- 输入音频为wav格式，api调用方式可参考如下范例:

```python
rec_result = inference_pipeline(input'asr_example_zh.wav')
```

- 若输入格式为文件wav.scp(注：文件名需要以.scp结尾)，可添加 output_dir 参数将识别结果写入文件中，api调用方式可参考如下范例:

```python
inference_pipeline(input="wav.scp", output_dir='./output_dir')
```
识别结果输出路径结构如下：

```sh
tree output_dir/
output_dir/
└── 1best_recog
    ├── score
    └── text

1 directory, 3 files
```
score：识别路径得分

text：语音识别结果文件


- 若输入音频为已解析的audio音频，api调用方式可参考如下范例：

```python
import soundfile

waveform, sample_rate = soundfile.read("asr_example_zh.wav")
rec_result = inference_pipeline(input=waveform)
```

- ASR、VAD、PUNC模型自由组合

可根据使用需求对VAD和PUNC标点模型进行自由组合，使用方式如下：
```python
inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch', model_revision="v2.0.4",
    vad_model='iic/speech_fsmn_vad_zh-cn-16k-common-pytorch', vad_model_revision="v2.0.4",
    punc_model='iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch', punc_model_revision="v2.0.4",
    # spk_model="iic/speech_campplus_sv_zh-cn_16k-common",
    # spk_model_revision="v2.0.2",
)
```
若不使用PUNC模型，可配置punc_model=""，或不传入punc_model参数，如需加入LM模型，可增加配置lm_model='damo/speech_transformer_lm_zh-cn-common-vocab8404-pytorch'，并设置lm_weight和beam_size参数。

## 基于FunASR进行推理

下面为快速上手教程，测试音频（[中文](https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/vad_example.wav)，[英文](https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_en.wav)）

### 可执行命令行
在命令行终端执行：

```shell
funasr ++model=paraformer-zh ++vad_model="fsmn-vad" ++punc_model="ct-punc" ++input=vad_example.wav
```

注：支持单条音频文件识别，也支持文件列表，列表为kaldi风格wav.scp：`wav_id   wav_path`

### python示例
#### 非实时语音识别
```python
from funasr import AutoModel
# paraformer-zh is a multi-functional asr model
# use vad, punc, spk or not as you need
model = AutoModel(model="paraformer-zh", model_revision="v2.0.4",
                  vad_model="fsmn-vad", vad_model_revision="v2.0.4",
                  punc_model="ct-punc-c", punc_model_revision="v2.0.4",
                  # spk_model="cam++", spk_model_revision="v2.0.2",
                  )
res = model.generate(input=f"{model.model_path}/example/asr_example.wav", 
            batch_size_s=300, 
            hotword='魔搭')
print(res)
```
注：`model_hub`：表示模型仓库，`ms`为选择modelscope下载，`hf`为选择huggingface下载。

#### 实时语音识别

```python
from funasr import AutoModel

chunk_size = [0, 10, 5] #[0, 10, 5] 600ms, [0, 8, 4] 480ms
encoder_chunk_look_back = 4 #number of chunks to lookback for encoder self-attention
decoder_chunk_look_back = 1 #number of encoder chunks to lookback for decoder cross-attention

model = AutoModel(model="paraformer-zh-streaming", model_revision="v2.0.4")

import soundfile
import os

wav_file = os.path.join(model.model_path, "example/asr_example.wav")
speech, sample_rate = soundfile.read(wav_file)
chunk_stride = chunk_size[1] * 960 # 600ms

cache = {}
total_chunk_num = int(len((speech)-1)/chunk_stride+1)
for i in range(total_chunk_num):
    speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
    is_final = i == total_chunk_num - 1
    res = model.generate(input=speech_chunk, cache=cache, is_final=is_final, chunk_size=chunk_size, encoder_chunk_look_back=encoder_chunk_look_back, decoder_chunk_look_back=decoder_chunk_look_back)
    print(res)
```

注：`chunk_size`为流式延时配置，`[0,10,5]`表示上屏实时出字粒度为`10*60=600ms`，未来信息为`5*60=300ms`。每次推理输入为`600ms`（采样点数为`16000*0.6=960`），输出为对应文字，最后一个语音片段输入需要设置`is_final=True`来强制输出最后一个字。

#### 语音端点检测（非实时）
```python
from funasr import AutoModel

model = AutoModel(model="fsmn-vad", model_revision="v2.0.4")

wav_file = f"{model.model_path}/example/asr_example.wav"
res = model.generate(input=wav_file)
print(res)
```

#### 语音端点检测（实时）
```python
from funasr import AutoModel

chunk_size = 200 # ms
model = AutoModel(model="fsmn-vad", model_revision="v2.0.4")

import soundfile

wav_file = f"{model.model_path}/example/vad_example.wav"
speech, sample_rate = soundfile.read(wav_file)
chunk_stride = int(chunk_size * sample_rate / 1000)

cache = {}
total_chunk_num = int(len((speech)-1)/chunk_stride+1)
for i in range(total_chunk_num):
    speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
    is_final = i == total_chunk_num - 1
    res = model.generate(input=speech_chunk, cache=cache, is_final=is_final, chunk_size=chunk_size)
    if len(res[0]["value"]):
        print(res)
```

#### 标点恢复
```python
from funasr import AutoModel

model = AutoModel(model="ct-punc", model_revision="v2.0.4")

res = model.generate(input="那今天的会就到这里吧 happy new year 明年见")
print(res)
```

#### 时间戳预测
```python
from funasr import AutoModel

model = AutoModel(model="fa-zh", model_revision="v2.0.4")

wav_file = f"{model.model_path}/example/asr_example.wav"
text_file = f"{model.model_path}/example/text.txt"
res = model.generate(input=(wav_file, text_file), data_type=("sound", "text"))
print(res)
```

更多详细用法（[示例](https://github.com/alibaba-damo-academy/FunASR/tree/main/examples/industrial_data_pretraining)）


## 微调

详细用法（[示例](https://github.com/alibaba-damo-academy/FunASR/tree/main/examples/industrial_data_pretraining)）

## 使用方式以及适用范围

运行范围
- 支持Linux-x86_64、Mac和Windows运行。

使用方式
- 直接推理：可以直接对输入音频进行解码，输出目标文字。

使用范围与目标场景
- 适合于离线语音识别场景


## 模型局限性以及可能的偏差

考虑到特征提取流程和工具以及训练工具差异，会对CER的数据带来一定的差异（<0.1%），推理GPU环境差异导致的RTF数值差异。



## 相关论文以及引用信息

```BibTeX
@inproceedings{radford2023robust,
  title={Robust speech recognition via large-scale weak supervision},
  author={Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya},
  booktitle={International Conference on Machine Learning},
  pages={28492--28518},
  year={2023},
  organization={PMLR}
}
```