shixian.shi
2023-10-13 9a0bc00e5fb2f892987216eafca8aeb140e17c6c
egs_modelscope/asr/TEMPLATE/README.md
@@ -1,12 +1,14 @@
([简体中文](./README_zh.md)|English)
# Speech Recognition
> **Note**:
> The modelscope pipeline supports all the models in [model zoo](https://alibaba-damo-academy.github.io/FunASR/en/modelscope_models.html#pretrained-models-on-modelscope) to inference and finetine. Here we take model of Paraformer and Paraformer-online as example to demonstrate the usage.
> **Note**:
> The modelscope pipeline supports all the models in [model zoo](https://alibaba-damo-academy.github.io/FunASR/en/model_zoo/modelscope_models.html#pretrained-models-on-modelscope) to inference and finetine. Here we take the typic models as examples to demonstrate the usage.
## Inference
### Quick start
#### [Paraformer model](https://www.modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary)
#### [Paraformer Model](https://www.modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary)
```python
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
@@ -19,30 +21,72 @@
rec_result = inference_pipeline(audio_in='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav')
print(rec_result)
```
#### [Paraformer-online model](https://www.modelscope.cn/models/damo/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online/summary)
#### [Paraformer-online Model](https://www.modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/summary)
##### Streaming Decoding
```python
inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='damo/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online',
    model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
    model_revision='v1.0.7',
    update_model=False,
    mode='paraformer_streaming'
    )
import soundfile
speech, sample_rate = soundfile.read("example/asr_example.wav")
param_dict = {"cache": dict(), "is_final": False}
chunk_stride = 7680# 480ms
# first chunk, 480ms
speech_chunk = speech[0:chunk_stride]
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
param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size,
              "encoder_chunk_look_back": encoder_chunk_look_back, "decoder_chunk_look_back": decoder_chunk_look_back}
chunk_stride = chunk_size[1] * 960 # 600ms、480ms
# first chunk, 600ms
speech_chunk = speech[0:chunk_stride]
rec_result = inference_pipeline(audio_in=speech_chunk, param_dict=param_dict)
print(rec_result)
# next chunk, 480ms
# next chunk, 600ms
speech_chunk = speech[chunk_stride:chunk_stride+chunk_stride]
rec_result = inference_pipeline(audio_in=speech_chunk, param_dict=param_dict)
print(rec_result)
```
##### Fake Streaming Decoding
```python
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
    model_revision='v1.0.7',
    update_model=False,
    mode="paraformer_fake_streaming"
)
audio_in='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav'
rec_result = inference_pipeline(audio_in=audio_in)
print(rec_result)
```
Full code of demo, please ref to [demo](https://github.com/alibaba-damo-academy/FunASR/discussions/241)
#### [UniASR model](https://www.modelscope.cn/models/damo/speech_UniASR_asr_2pass-zh-cn-8k-common-vocab3445-pytorch-online/summary)
There are three decoding mode for UniASR model(`fast`、`normal`、`offline`), for more model detailes, please refer to [docs](https://www.modelscope.cn/models/damo/speech_UniASR_asr_2pass-zh-cn-8k-common-vocab3445-pytorch-online/summary)
#### [Paraformer-contextual Model](https://www.modelscope.cn/models/damo/speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404/summary)
```python
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
param_dict = dict()
# param_dict['hotword'] = "https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/hotword.txt"
param_dict['hotword']="邓郁松 王颖春 王晔君"
inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model="damo/speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404",
    param_dict=param_dict)
rec_result = inference_pipeline(audio_in='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_hotword.wav')
print(rec_result)
```
#### [UniASR Model](https://www.modelscope.cn/models/damo/speech_UniASR_asr_2pass-zh-cn-8k-common-vocab3445-pytorch-online/summary)
There are three decoding mode for UniASR model(`fast`、`normal`、`offline`), for more model details, please refer to [docs](https://www.modelscope.cn/models/damo/speech_UniASR_asr_2pass-zh-cn-8k-common-vocab3445-pytorch-online/summary)
```python
decoding_model = "fast" # "fast"、"normal"、"offline"
inference_pipeline = pipeline(
@@ -53,47 +97,89 @@
rec_result = inference_pipeline(audio_in='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav')
print(rec_result)
```
The decoding mode of `fast` and `normal`
The decoding mode of `fast` and `normal` is fake streaming, which could be used for evaluating of recognition accuracy.
Full code of demo, please ref to [demo](https://github.com/alibaba-damo-academy/FunASR/discussions/151)
#### [Paraformer-Spk](https://modelscope.cn/models/damo/speech_paraformer-large-vad-punc-spk_asr_nat-zh-cn/summary)
This model allows user to get recognition results which contain speaker info of each sentence. Refer to [CAM++](https://modelscope.cn/models/damo/speech_campplus_speaker-diarization_common/summary) for detailed information about speaker diarization model.
```python
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_speaker_demo.wav'
    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model='damo/speech_paraformer-large-vad-punc-spk_asr_nat-zh-cn',
        model_revision='v0.0.2',
        vad_model='damo/speech_fsmn_vad_zh-cn-16k-common-pytorch',
        punc_model='damo/punc_ct-transformer_cn-en-common-vocab471067-large',
        output_dir=output_dir,
    )
    rec_result = inference_pipeline(audio_in=audio_in, batch_size_token=5000, batch_size_token_threshold_s=40, max_single_segment_time=6000)
    print(rec_result)
```
#### [RNN-T-online model]()
Undo
#### API-reference
##### define pipeline
#### [MFCCA Model](https://www.modelscope.cn/models/NPU-ASLP/speech_mfcca_asr-zh-cn-16k-alimeeting-vocab4950/summary)
For more model details, please refer to [docs](https://www.modelscope.cn/models/NPU-ASLP/speech_mfcca_asr-zh-cn-16k-alimeeting-vocab4950/summary)
```python
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='NPU-ASLP/speech_mfcca_asr-zh-cn-16k-alimeeting-vocab4950',
    model_revision='v3.0.0'
)
rec_result = inference_pipeline(audio_in='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav')
print(rec_result)
```
### API-reference
#### Define pipeline
- `task`: `Tasks.auto_speech_recognition`
- `model`: model name in [model zoo](https://alibaba-damo-academy.github.io/FunASR/en/modelscope_models.html#pretrained-models-on-modelscope), or model path in local disk
- `ngpu`: 1 (Defalut), decoding on GPU. If ngpu=0, decoding on CPU
- `ncpu`: 1 (Defalut), sets the number of threads used for intraop parallelism on CPU
- `output_dir`: None (Defalut), the output path of results if set
- `batch_size`: 1 (Defalut), batch size when decoding
##### infer pipeline
- `audio_in`: the input to decode, which could be:
- `model`: model name in [model zoo](https://alibaba-damo-academy.github.io/FunASR/en/model_zoo/modelscope_models.html#pretrained-models-on-modelscope), or model path in local disk
- `ngpu`: `1` (Default), decoding on GPU. If ngpu=0, decoding on CPU
- `ncpu`: `1` (Default), sets the number of threads used for intraop parallelism on CPU
- `output_dir`: `None` (Default), the output path of results if set
- `batch_size`: `1` (Default), batch size when decoding
#### Infer pipeline
- `audio_in`: the input to decode, which could be:
  - wav_path, `e.g.`: asr_example.wav,
  - pcm_path, `e.g.`: asr_example.pcm,
  - pcm_path, `e.g.`: asr_example.pcm,
  - audio bytes stream, `e.g.`: bytes data from a microphone
  - audio sample point,`e.g.`: `audio, rate = soundfile.read("asr_example_zh.wav")`, the dtype is numpy.ndarray or torch.Tensor
  - wav.scp, kaldi style wav list (`wav_id \t wav_path``), `e.g.`:
  ```cat wav.scp
  - wav.scp, kaldi style wav list (`wav_id \t wav_path`), `e.g.`:
  ```text
  asr_example1  ./audios/asr_example1.wav
  asr_example2  ./audios/asr_example2.wav
  ```
  In this case of `wav.scp` input, `output_dir` must be set to save the output results
- `audio_fs`: audio sampling rate, only set when audio_in is pcm audio
- `output_dir`: None (Defalut), the output path of results if set
- `output_dir`: None (Default), the output path of results if set
### Inference with multi-thread CPUs or multi GPUs
FunASR also offer recipes [infer.sh](https://github.com/alibaba-damo-academy/FunASR/blob/main/egs_modelscope/asr/TEMPLATE/infer.sh) to decode with multi-thread CPUs, or multi GPUs.
FunASR also offer recipes [egs_modelscope/asr/TEMPLATE/infer.sh](https://github.com/alibaba-damo-academy/FunASR/blob/main/egs_modelscope/asr/TEMPLATE/infer.sh) to decode with multi-thread CPUs, or multi GPUs.
- Setting parameters in `infer.sh`
    - <strong>model:</strong> # model name on ModelScope
    - <strong>data_dir:</strong> # the dataset dir needs to include `${data_dir}/wav.scp`. If `${data_dir}/text` is also exists, CER will be computed
    - <strong>output_dir:</strong> # result dir
    - <strong>batch_size:</strong> # batchsize of inference
    - <strong>gpu_inference:</strong> # whether to perform gpu decoding, set false for cpu decoding
    - <strong>gpuid_list:</strong> # set gpus, e.g., gpuid_list="0,1"
    - <strong>njob:</strong> # the number of jobs for CPU decoding, if `gpu_inference`=false, use CPU decoding, please set `njob`
#### Settings of `infer.sh`
- `model`: model name in [model zoo](https://alibaba-damo-academy.github.io/FunASR/en/model_zoo/modelscope_models.html#pretrained-models-on-modelscope), or model path in local disk
- `data_dir`: the dataset dir needs to include `wav.scp`. If `${data_dir}/text` is also exists, CER will be computed
- `output_dir`: output dir of the recognition results
- `batch_size`: `64` (Default), batch size of inference on gpu
- `gpu_inference`: `true` (Default), whether to perform gpu decoding, set false for CPU inference
- `gpuid_list`: `0,1` (Default), which gpu_ids are used to infer
- `njob`: only used for CPU inference (`gpu_inference`=`false`), `64` (Default), the number of jobs for CPU decoding
- `checkpoint_dir`: only used for infer finetuned models, the path dir of finetuned models
- `checkpoint_name`: only used for infer finetuned models, `valid.cer_ctc.ave.pb` (Default), which checkpoint is used to infer
- `decoding_mode`: `normal` (Default), decoding mode for UniASR model(fast、normal、offline)
- `hotword_txt`: `None` (Default), hotword file for contextual paraformer model(the hotword file name ends with .txt")
- Decode with multi GPUs:
#### Decode with multi GPUs:
```shell
    bash infer.sh \
    --model "damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch" \
@@ -103,7 +189,7 @@
    --gpu_inference true \
    --gpuid_list "0,1"
```
- Decode with multi-thread CPUs:
#### Decode with multi-thread CPUs:
```shell
    bash infer.sh \
    --model "damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch" \
@@ -113,7 +199,7 @@
    --njob 64
```
- Results
#### Results
The decoding results can be found in `$output_dir/1best_recog/text.cer`, which includes recognition results of each sample and the CER metric of the whole test set.
@@ -126,15 +212,19 @@
[finetune.py](https://github.com/alibaba-damo-academy/FunASR/blob/main/egs_modelscope/asr/TEMPLATE/finetune.py)
```python
import os
from modelscope.metainfo import Trainers
from modelscope.trainers import build_trainer
from modelscope.msdatasets.audio.asr_dataset import ASRDataset
from funasr.datasets.ms_dataset import MsDataset
from funasr.utils.modelscope_param import modelscope_args
def modelscope_finetune(params):
    if not os.path.exists(params.output_dir):
        os.makedirs(params.output_dir, exist_ok=True)
    # dataset split ["train", "validation"]
    ds_dict = ASRDataset.load(params.data_path, namespace='speech_asr')
    ds_dict = MsDataset.load(params.data_path)
    kwargs = dict(
        model=params.model,
        data_dir=ds_dict,
@@ -142,21 +232,32 @@
        work_dir=params.output_dir,
        batch_bins=params.batch_bins,
        max_epoch=params.max_epoch,
        lr=params.lr)
        lr=params.lr,
        mate_params=params.param_dict)
    trainer = build_trainer(Trainers.speech_asr_trainer, default_args=kwargs)
    trainer.train()
if __name__ == '__main__':
    from funasr.utils.modelscope_param import modelscope_args
    params = modelscope_args(model="damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch")
    params.output_dir = "./checkpoint"                      # 模型保存路径
    params.data_path = "speech_asr_aishell1_trainsets"      # 数据路径,可以为modelscope中已上传数据,也可以是本地数据
    params.dataset_type = "small"                           # 小数据量设置small,若数据量大于1000小时,请使用large
    params.batch_bins = 2000                                # batch size,如果dataset_type="small",batch_bins单位为fbank特征帧数,如果dataset_type="large",batch_bins单位为毫秒,
    params.max_epoch = 50                                   # 最大训练轮数
    params.lr = 0.00005                                     # 设置学习率
    params = modelscope_args(model="damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch", data_path="./data")
    params.output_dir = "./checkpoint"              # m模型保存路径
    params.data_path = "speech_asr_aishell1_trainsets"            # 数据路径
    params.dataset_type = "small"                   # 小数据量设置small,若数据量大于1000小时,请使用large
    params.batch_bins = 2000                       # batch size,如果dataset_type="small",batch_bins单位为fbank特征帧数,如果dataset_type="large",batch_bins单位为毫秒,
    params.max_epoch = 20                           # 最大训练轮数
    params.lr = 0.00005                             # 设置学习率
    init_param = []                                 # 初始模型路径,默认加载modelscope模型初始化,例如: ["checkpoint/20epoch.pb"]
    freeze_param = []                               # 模型参数freeze, 例如: ["encoder"]
    ignore_init_mismatch = True                     # 是否忽略模型参数初始化不匹配
    use_lora = False                                # 是否使用lora进行模型微调
    params.param_dict = {"init_param":init_param, "freeze_param": freeze_param, "ignore_init_mismatch": ignore_init_mismatch}
    if use_lora:
        enable_lora = True
        lora_bias = "all"
        lora_params = {"lora_list":['q','v'], "lora_rank":8, "lora_alpha":16, "lora_dropout":0.1}
        lora_config = {"enable_lora": enable_lora, "lora_bias": lora_bias, "lora_params": lora_params}
        params.param_dict.update(lora_config)
    modelscope_finetune(params)
```
@@ -167,12 +268,31 @@
### Finetune with your data
- Modify finetune training related parameters in [finetune.py](https://github.com/alibaba-damo-academy/FunASR/blob/main/egs_modelscope/asr/TEMPLATE/finetune.py)
    - <strong>output_dir:</strong> # result dir
    - <strong>data_dir:</strong> # the dataset dir needs to include files: `train/wav.scp`, `train/text`; `validation/wav.scp`, `validation/text`
    - <strong>dataset_type:</strong> # for dataset larger than 1000 hours, set as `large`, otherwise set as `small`
    - <strong>batch_bins:</strong> # batch size. For dataset_type is `small`, `batch_bins` indicates the feature frames. For dataset_type is `large`, `batch_bins` indicates the duration in ms
    - <strong>max_epoch:</strong> # number of training epoch
    - <strong>lr:</strong> # learning rate
    - `output_dir`: result dir
    - `data_dir`: the dataset dir needs to include files: `train/wav.scp`, `train/text`; `validation/wav.scp`, `validation/text`
    - `dataset_type`: for dataset larger than 1000 hours, set as `large`, otherwise set as `small`
    - `batch_bins`: batch size. For dataset_type is `small`, `batch_bins` indicates the feature frames. For dataset_type is `large`, `batch_bins` indicates the duration in ms
    - `max_epoch`: number of training epoch
    - `lr`: learning rate
    - `init_param`: `[]`(Default), init model path, load modelscope model initialization by default. For example: ["checkpoint/20epoch.pb"]
    - `freeze_param`: `[]`(Default), Freeze model parameters. For example:["encoder"]
    - `ignore_init_mismatch`: `True`(Default), Ignore size mismatch when loading pre-trained model
    - `use_lora`: `False`(Default), Fine-tuning model use lora, more detail please refer to [LORA](https://arxiv.org/pdf/2106.09685.pdf)
- Training data formats:
```sh
cat ./example_data/text
BAC009S0002W0122 而 对 楼 市 成 交 抑 制 作 用 最 大 的 限 购
BAC009S0002W0123 也 成 为 地 方 政 府 的 眼 中 钉
english_example_1 hello world
english_example_2 go swim 去 游 泳
cat ./example_data/wav.scp
BAC009S0002W0122 /mnt/data/wav/train/S0002/BAC009S0002W0122.wav
BAC009S0002W0123 /mnt/data/wav/train/S0002/BAC009S0002W0123.wav
english_example_1 /mnt/data/wav/train/S0002/english_example_1.wav
english_example_2 /mnt/data/wav/train/S0002/english_example_2.wav
```
- Then you can run the pipeline to finetune with:
```shell
@@ -183,14 +303,29 @@
CUDA_VISIBLE_DEVICES=1,2 python -m torch.distributed.launch --nproc_per_node 2 finetune.py > log.txt 2>&1
```
## Inference with your finetuned model
- Modify inference related parameters in [infer_after_finetune.py](https://github.com/alibaba-damo-academy/FunASR/blob/main/egs_modelscope/asr/TEMPLATE/infer_after_finetune.py)
    - <strong>modelscope_model_name: </strong> # model name on ModelScope
    - <strong>output_dir:</strong> # result dir
    - <strong>data_dir:</strong> # the dataset dir needs to include `test/wav.scp`. If `test/text` is also exists, CER will be computed
    - <strong>decoding_model_name:</strong> # set the checkpoint name for decoding, e.g., `valid.cer_ctc.ave.pb`
    - <strong>batch_size:</strong> # batchsize of inference
- Then you can run the pipeline to finetune with:
```python
    python infer_after_finetune.py
- Setting parameters in [egs_modelscope/asr/TEMPLATE/infer.sh](https://github.com/alibaba-damo-academy/FunASR/blob/main/egs_modelscope/asr/TEMPLATE/infer.sh) is the same with [docs](https://github.com/alibaba-damo-academy/FunASR/tree/main/egs_modelscope/asr/TEMPLATE#inference-with-multi-thread-cpus-or-multi-gpus), `model` is the model name from modelscope, which you finetuned.
- Decode with multi GPUs:
```shell
    bash infer.sh \
    --model "damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch" \
    --data_dir "./data/test" \
    --output_dir "./results" \
    --batch_size 64 \
    --gpu_inference true \
    --gpuid_list "0,1" \
    --checkpoint_dir "./checkpoint" \
    --checkpoint_name "valid.cer_ctc.ave.pb"
```
- Decode with multi-thread CPUs:
```shell
    bash infer.sh \
    --model "damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch" \
    --data_dir "./data/test" \
    --output_dir "./results" \
    --gpu_inference false \
    --njob 64 \
    --checkpoint_dir "./checkpoint" \
    --checkpoint_name "valid.cer_ctc.ave.pb"
```