qwenaudio qwenaudiochat (#1433)

zhifu gao

2024-03-05 f2406e2d61218c848bfd6da933c36956a9b0a5aa

qwenaudio qwenaudiochat (#1433)

 README.md

@@ -27,7 +27,8 @@

<a name="whats-new"></a>
## What's new:
- 2024/03/05：Added support for the Whisper-large-v3 model, a multitasking model that can perform multilingual speech recognition, speech translation, and language identification. It can be downloaded from the[modelscope](https://www.modelscope.cn/models/iic/Whisper-large-v3/summary), and [openai](https://github.com/alibaba-damo-academy/FunASR/tree/main/examples/industrial_data_pretraining/whisper).
- 2024/03/05：Added the Qwen-Audio and Qwen-Audio-Chat large-scale audio-text multimodal models, which have topped multiple audio domain leaderboards. These models support speech dialogue, [usage](examples/industrial_data_pretraining/qwen_audio).
- 2024/03/05：Added support for the Whisper-large-v3 model, a multitasking model that can perform multilingual speech recognition, speech translation, and language identification. It can be downloaded from the[modelscope](examples/industrial_data_pretraining/whisper/demo.py), and [openai](examples/industrial_data_pretraining/whisper/demo_from_openai.py).
- 2024/03/05: Offline File Transcription Service 4.4, Offline File Transcription Service of English 1.5，Real-time Transcription Service 1.9 released，docker image supports ARM64 platform, update modelscope；([docs](runtime/readme.md))
- 2024/01/30：funasr-1.0 has been released ([docs](https://github.com/alibaba-damo-academy/FunASR/discussions/1319))
- 2024/01/30：emotion recognition models are new supported. [model link](https://www.modelscope.cn/models/iic/emotion2vec_base_finetuned/summary), modified from [repo](https://github.com/ddlBoJack/emotion2vec).
@@ -83,6 +84,8 @@
|                                       cam++ <br> ( [⭐](https://modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common/summary) [🤗](https://huggingface.co/funasr/campplus) )                                        |           speaker verification/diarization            |            5000 hours            |    7.2M    | 
|                                                  Whisper-large-v2 <br> ([⭐](https://www.modelscope.cn/models/iic/speech_whisper-large_asr_multilingual/summary)  [🍀](https://github.com/openai/whisper) )                                                  |  speech recognition, with timestamps, non-streaming   |          multilingual            |    1.5G    |
|                                                Whisper-large-v3 <br> ([⭐](https://www.modelscope.cn/models/iic/Whisper-large-v3/summary)  [🍀](https://github.com/openai/whisper) )                                                 |  speech recognition, with timestamps, non-streaming   |          multilingual            |    1.5G    |
|                                         Qwen-Audio <br> ([⭐](examples/industrial_data_pretraining/qwen_audio/demo.py)  [🤗](https://huggingface.co/Qwen/Qwen-Audio) )                                         |      audio-text multimodal models (pretraining)       |     multilingual      |  8B  |
|                   Qwen-Audio-Chat <br> ([⭐](examples/industrial_data_pretraining/qwen_audio/demo_chat.py)  [🤗](https://huggingface.co/Qwen/Qwen-Audio-Chat) )                                                |          audio-text multimodal models (chat)          |     multilingual      |  8B  |




 README_zh.md

@@ -29,7 +29,8 @@

<a name="最新动态"></a>
## 最新动态
- 2024/03/05：新增加Whisper-large-v3模型支持，多语言语音识别/翻译/语种识别，支持从[modelscope](https://www.modelscope.cn/models/iic/Whisper-large-v3/summary)仓库下载，也支持从[openai](https://github.com/alibaba-damo-academy/FunASR/tree/main/examples/industrial_data_pretraining/whisper)仓库下载模型。
- 2024/03/05：新增加Qwen-Audio与Qwen-Audio-Chat音频文本模态大模型，在多个音频领域测试榜单刷榜，中支持语音对话，详细用法见 [示例](examples/industrial_data_pretraining/qwen_audio)。
- 2024/03/05：新增加Whisper-large-v3模型支持，多语言语音识别/翻译/语种识别，支持从 [modelscope](examples/industrial_data_pretraining/whisper/demo.py)仓库下载，也支持从 [openai](examples/industrial_data_pretraining/whisper/demo_from_openai.py)仓库下载模型。
- 2024/03/05: 中文离线文件转写服务 4.4、英文离线文件转写服务 1.5、中文实时语音听写服务 1.9 发布，docker镜像支持arm64平台，升级modelscope版本；详细信息参阅([部署文档](runtime/readme_cn.md))
- 2024/01/30：funasr-1.0发布，更新说明[文档](https://github.com/alibaba-damo-academy/FunASR/discussions/1319)
- 2024/01/30：新增加情感识别 [模型链接](https://www.modelscope.cn/models/iic/emotion2vec_base_finetuned/summary)，原始模型 [repo](https://github.com/ddlBoJack/emotion2vec).
@@ -74,7 +75,7 @@


|                                                                                                     模型名字                                                                                                      |      任务详情       |     训练数据     | 参数量  | 
|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:---------------:|:------------:|:----:|
|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:------------------:|:------------:|:----:|
|    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-tp) )    | 语音识别，带时间戳输出，非实时 |  60000小时，中文  | 220M |
| 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) ) |     语音识别，实时     |  60000小时，中文  | 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) )         |    语音识别，非实时     |  50000小时，英文  | 220M |
@@ -83,9 +84,9 @@
|                            fsmn-vad <br> ( [⭐](https://modelscope.cn/models/damo/speech_fsmn_vad_zh-cn-16k-common-pytorch/summary) [🤗](https://huggingface.co/funasr/fsmn-vad) )                             |    语音端点检测，实时    | 5000小时，中文与英文 | 0.4M | 
|                              fa-zh <br> ( [⭐](https://modelscope.cn/models/damo/speech_timestamp_prediction-v1-16k-offline/summary) [🤗](https://huggingface.co/funasr/fa-zh) )                               |    字级别时间戳预测     |  50000小时，中文  | 38M  |
|                                 cam++ <br> ( [⭐](https://modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common/summary) [🤗](https://huggingface.co/funasr/campplus) )                                 |    说话人确认/分割     |    5000小时    | 7.2M | 
|                           Whisper-large-v2 <br> ([⭐](https://www.modelscope.cn/models/iic/speech_whisper-large_asr_multilingual/summary)  [🍀](https://github.com/openai/whisper) )                           | 语音识别，带时间戳输出，非实时 |     多语言      |  1G  |
|                         Whisper-large-v3 <br> ([⭐](https://www.modelscope.cn/models/iic/Whisper-large-v3/summary)  [🍀](https://github.com/openai/whisper) )                          | 语音识别，带时间戳输出，非实时 |     多语言      |  1G  |

|                                         Qwen-Audio <br> ([⭐](examples/industrial_data_pretraining/qwen_audio/demo.py)  [🤗](https://huggingface.co/Qwen/Qwen-Audio) )                                         |  音频文本多模态大模型（预训练）   |     多语言      |  8B  |
|                   Qwen-Audio-Chat <br> ([⭐](examples/industrial_data_pretraining/qwen_audio/demo_chat.py)  [🤗](https://huggingface.co/Qwen/Qwen-Audio-Chat) )                                                | 音频文本多模态大模型（chat版本） |     多语言      |  8B  |

<a name="快速开始"></a>
## 快速开始

 examples/industrial_data_pretraining/llm_asr/conf/template.yaml

New file
@@ -0,0 +1,89 @@
# This is an example that demonstrates how to configure a model file.
# You can modify the configuration according to your own requirements.

# to print the register_table:
# from funasr.register import tables
# tables.print()

# network architecture
model: LLMASR
model_conf:
    lsm_weight: 0.1     # label smoothing option
    length_normalized_loss: true

# encoder
encoder: WhisperWarp
encoder_conf:
    hub: funasr
    init_param_path: "/nfs/maziyang.mzy/models/Whisper-large-v2"
    freeze: true

llm: Vicuna
llm_conf:
  hub: hf
  init_param_path: "/nfs/maziyang.mzy/models/vicuna-7b-v1.5"
  freeze: true

adaptor: Linear
adaptor_conf:
  downsample_rate: 5
  llm_dim: 4096
  encoder_dim: 512

# frontend related
frontend: WhisperFrontend
frontend_conf:
    fs: 16000
    whisper_model: large
    do_pad_trim: true


specaug: SpecAugLFR
specaug_conf:
    apply_time_warp: false
    time_warp_window: 5
    time_warp_mode: bicubic
    apply_freq_mask: true
    freq_mask_width_range:
    - 0
    - 30
    lfr_rate: 6
    num_freq_mask: 1
    apply_time_mask: true
    time_mask_width_range:
    - 0
    - 12
    num_time_mask: 1

train_conf:
  accum_grad: 1
  grad_clip: 5
  max_epoch: 150
  keep_nbest_models: 10
  log_interval: 10

optim: adamw
optim_conf:
   lr: 0.0001
   weight_decay: 0.000001
scheduler: warmuplr
scheduler_conf:
   warmup_steps: 1500

dataset: AudioLLMDataset
dataset_conf:
    index_ds: IndexDSJsonl
    batch_sampler: RankFullLocalShuffleBatchSampler
    batch_type: example # example or length
    batch_size: 8 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
    max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
    buffer_size: 500
    shuffle: True
    num_workers: 4
    preprocessor_text: TextPreprocessRemovePunctuation

tokenizer: HuggingfaceTokenizer
tokenizer_conf:
  unk_symbol: <unk>
  init_param_path: "/nfs/maziyang.mzy/models/vicuna-7b-v1.5"


 examples/industrial_data_pretraining/llm_asr/demo_infer.sh

New file
@@ -0,0 +1,14 @@
# Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
#  MIT License  (https://opensource.org/licenses/MIT)



python -m funasr.bin.inference \
--config-path="/root/FunASR/examples/aishell/llm_asr_nar/conf" \
--config-name="template.yaml" \
++init_param="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3/model.pt.ep38" \
++input="/nfs/beinian.lzr/workspace/datasets/data/16k/opendata/aishell1/dev/wav/S0724/BAC009S0724W0121.wav" \
++scope_map="encoder.model,audio_encoder,encoder_projector,adaptor" \
++output_dir="./outputs/debug" \
++device="cpu" \


 examples/industrial_data_pretraining/llm_asr/demo_train_or_finetune.sh

New file
@@ -0,0 +1,47 @@
# Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
#  MIT License  (https://opensource.org/licenses/MIT)


# which gpu to train or finetune
export CUDA_VISIBLE_DEVICES="0"
gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')

# data dir, which contains: train.json, val.json, tokens.jsonl/tokens.txt, am.mvn
#data_dir="/Users/zhifu/funasr1.0/data/list"

## generate jsonl from wav.scp and text.txt
#python -m funasr.datasets.audio_datasets.scp2jsonl \
#++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \
#++data_type_list='["source", "target"]' \
#++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl

train_data="/nfs/zhifu.gzf/data/datalist/aishell1_aishell2_wav_speech_llm_train_data_del_tail500.json"
val_data="/nfs/zhifu.gzf/data/datalist/aishell1_aishell2_wav_speech_llm_train_data_tail500.json"

# exp output dir
output_dir="/Users/zhifu/exp"
log_file="${output_dir}/log.txt"

workspace=`pwd`
config="template.yaml"

init_param="${output_dir}/model.pt"

mkdir -p ${output_dir}
echo "log_file: ${log_file}"

torchrun \
--nnodes 1 \
--nproc_per_node ${gpu_num} \
../../../funasr/bin/train.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
++train_data_set_list="${train_data}" \
++valid_data_set_list="${val_data}" \
++dataset_conf.batch_size=2 \
++dataset_conf.batch_type="example" \
++dataset_conf.num_workers=0 \
++train_conf.max_epoch=11 \
++optim_conf.lr=0.0002 \
++init_param="${init_param}" \
++output_dir="${output_dir}" &> ${log_file}

 funasr/models/qwen_audio/model.py

@@ -20,6 +20,11 @@
@tables.register("model_classes", "QwenAudio")
@tables.register("model_classes", "QwenAudioWarp")
class QwenAudioWarp(nn.Module):
    """
    Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models
    https://arxiv.org/abs/2311.07919
    Modified from https://github.com/QwenLM/Qwen-Audio
    """
    def __init__(self, *args, **kwargs):
        super().__init__()

@@ -72,6 +77,11 @@
@tables.register("model_classes", "QwenAudioChatWarp")
class QwenAudioChatWarp(nn.Module):
    def __init__(self, *args, **kwargs):
        """
        Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models
        https://arxiv.org/abs/2311.07919
        Modified from https://github.com/QwenLM/Qwen-Audio
        """
        super().__init__()
        
        model_or_path = kwargs.get("model_path", "QwenAudio")

 funasr/tokenizer/hf_tokenizer.py

@@ -2,7 +2,8 @@
try:
    from transformers import AutoTokenizer
except:
    print("If you want to use hugging, please `pip install -U transformers`")
    # print("If you want to use hugging, please `pip install -U transformers`")
    pass

from funasr.register import tables


 funasr/tokenizer/whisper_tokenizer.py

@@ -2,7 +2,7 @@
try:
    from whisper.tokenizer import get_tokenizer
except:
    print("If you want to use hugging, please `pip install -U transformers`")
    print("Notice: If you want to use whisper, please `pip install -U openai-whisper`")

from funasr.register import tables