新一代kaldi-icefall环境配置与aishell实践

新一代kaldi-icefall环境配置与实践

代码来源:[k2-fsa/icefall (github.com)](https://github.com/kaldi-asr/kaldi)
官网文档:https://k2-fsa.github.io/icefall/

一.环境配置

1.安装CUDA

# 1.创建conda虚拟环境
conda create -n k2 python=3.8
conda activate k2
# 2.安装CUDA cudnn torch torchaudio
# CUDA安装方法参考https://blog.youkuaiyun.com/weixin_46560570/article/details/140754242?spm=1001.2014.3001.5501
# 我的CUDA版本为11.6
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
# 安装cudnn
conda install cudnn

2.安装k2

# 根据你的cuda,torch以及python版本选择预编译的k2 :https://k2-fsa.github.io/k2/cuda.html
wget https://huggingface.co/csukuangfj/k2/resolve/main/ubuntu-cuda/k2-1.24.4.dev20240223+cuda11.6.torch1.12.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
pip install ./k2-1.24.4.dev20240223+cuda11.6.torch1.12.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

3.安装lhotse

pip install git+https://gitee.com/chengsili/lhotse

4.安装其他依赖

cd icefall
pip install -r requirements.txt

5.测试

# 数据准备
export PYTHONPATH=/s6home/lnj524/module/icefall:$PYTHONPATH
cd egs/yesno/ASR
./prepare.sh
# 训练
 export CUDA_VISIBLE_DEVICES=0,1
 ./tdnn/train.py
# 解码
./tdnn/decode.py

二.Aishell数据集

官方文档:Conformer CTC — icefall 0.1 documentation (k2-fsa.github.io)

1.数据准备

cd egs/aishell/ASR
# 修改prepare.sh 
# 将perturb_speed=false关闭数据增强
# dl_dir改为数据集存放地址
# aishell_lm可自行下载,放置在dl_dir/lm目录下
./prepare.sh

2.训练

# 若报错
Traceback (most recent call last):
  File "./conformer_ctc/decode.py", line 29, in <module>
    from asr_datamodule import AishellAsrDataModule
  File "/s6home/lnj524/module/icefall/egs/aishell/ASR/conformer_ctc/asr_datamodule.py", line 39, in <module>
    from icefall.utils import str2bool
ModuleNotFoundError: No module named 'icefall'
#解决方法
export PYTHONPATH=/s6home/lnj524/module/icefall:$PYTHONPATH
cd egs/aishell/ASR
export PYTHONPATH=/s6home/lnj524/module/icefall:$PYTHONPATH
export CUDA_VISIBLE_DEVICES=0,2,5,7
./conformer_ctc/train.py --world-size 4 --num-epochs 50 

3.tensorboard可视化

# 将conformer_ctc/exp/目录下tensorboard文件夹拉取到本机
tensorboard --logdir tensorboard/ --port 12598 --bind_all

在这里插入图片描述

4.解码

cd egs/aishell/ASR
export PYTHONPATH=/s6home/lnj524/module/icefall:$PYTHONPATH
export CUDA_VISIBLE_DEVICES=7
./conformer_ctc/decode.py --method attention-decoder --nbest-scale 0.5 --avg 5 --max-duration 60

2024-08-18 11:54:32,456 INFO [decode.py:474] Decoding started
2024-08-18 11:54:32,456 INFO [decode.py:475] {'subsampling_factor': 4, 'feature_dim': 80, 'nhead': 4, 'attention_dim': 512, 'num_encoder_layers': 12, 'num_decoder_layers': 6, 'vgg_frontend': False, 'use_feat_batchnorm': True, 'search_beam': 20, 'output_beam': 7, 'min_active_states': 30, 'max_active_states': 10000, 'use_double_scores': True, 'env_info': {'k2-version': '1.24.4', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'ff1d435a8d3c4eaa15828a84a7240678a70539a7', 'k2-git-date': 'Fri Feb 23 01:48:38 2024', 'lhotse-version': '1.27.0.dev+git.66b95ba.clean', 'torch-version': '1.12.1', 'torch-cuda-available': True, 'torch-cuda-version': '11.6', 'python-version': '3.8', 'icefall-git-branch': 'master', 'icefall-git-sha1': '3b257dd5-dirty', 'icefall-git-date': 'Thu Jul 25 16:46:24 2024', 'icefall-path': '/s6home/lnj524/module/icefall', 'k2-path': '/s6home/lnj524/miniconda3/envs/k2/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/s6home/lnj524/miniconda3/envs/k2/lib/python3.8/site-packages/lhotse/__init__.py', 'hostname': 'xju-aslp5', 'IP address': '127.0.1.1'}, 'epoch': 49, 'avg': 5, 'method': 'attention-decoder', 'num_paths': 100, 'nbest_scale': 0.5, 'exp_dir': PosixPath('conformer_ctc/exp'), 'lang_dir': PosixPath('data/lang_char'), 'lm_dir': PosixPath('data/lm'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 30, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True}
2024-08-18 11:54:32,497 INFO [lexicon.py:168] Loading pre-compiled data/lang_char/Linv.pt
2024-08-18 11:54:32,504 INFO [decode.py:485] device: cuda:0
2024-08-18 11:54:34,727 INFO [decode.py:533] averaging ['conformer_ctc/exp/epoch-45.pt', 'conformer_ctc/exp/epoch-46.pt', 'conformer_ctc/exp/epoch-47.pt', 'conformer_ctc/exp/epoch-48.pt', 'conformer_ctc/exp/epoch-49.pt']
2024-08-18 11:54:38,316 INFO [decode.py:540] Number of model parameters: 115125888
2024-08-18 11:54:38,316 INFO [asr_datamodule.py:378] About to get test cuts
2024-08-18 11:54:38,318 INFO [asr_datamodule.py:343] About to create test dataset
2024-08-18 11:54:42,084 INFO [decode.py:404] batch 0/?, cuts processed until now is 4
2024-08-18 11:55:31,261 INFO [decode.py:404] batch 100/?, cuts processed until now is 542
2024-08-18 11:56:19,738 INFO [decode.py:404] batch 200/?, cuts processed until now is 1060
2024-08-18 11:57:09,801 INFO [decode.py:404] batch 300/?, cuts processed until now is 1553
2024-08-18 11:58:00,809 INFO [decode.py:404] batch 400/?, cuts processed until now is 2041
2024-08-18 11:58:50,981 INFO [decode.py:404] batch 500/?, cuts processed until now is 2506
2024-08-18 11:59:41,018 INFO [decode.py:404] batch 600/?, cuts processed until now is 3015
2024-08-18 12:00:30,553 INFO [decode.py:404] batch 700/?, cuts processed until now is 3519
2024-08-18 12:01:20,724 INFO [decode.py:404] batch 800/?, cuts processed until now is 4014
2024-08-18 12:02:09,583 INFO [decode.py:404] batch 900/?, cuts processed until now is 4491
2024-08-18 12:02:58,389 INFO [decode.py:404] batch 1000/?, cuts processed until now is 5002
2024-08-18 12:03:47,875 INFO [decode.py:404] batch 1100/?, cuts processed until now is 5531
2024-08-18 12:04:35,366 INFO [decode.py:404] batch 1200/?, cuts processed until now is 6022
2024-08-18 12:05:25,122 INFO [decode.py:404] batch 1300/?, cuts processed until now is 6523
2024-08-18 12:06:12,023 INFO [decode.py:404] batch 1400/?, cuts processed until now is 7101
2024-08-18 12:10:24,922 INFO [decode.py:458] 
For test, CER of different settings are:
ngram_lm_scale_0.3_attention_scale_0.9	5.28	best for test
ngram_lm_scale_0.5_attention_scale_1.0	5.28
ngram_lm_scale_0.5_attention_scale_1.1	5.28
ngram_lm_scale_0.3_attention_scale_0.6	5.29
ngram_lm_scale_0.3_attention_scale_0.7	5.29
ngram_lm_scale_0.3_attention_scale_1.0	5.29
ngram_lm_scale_0.3_attention_scale_1.1	5.29
ngram_lm_scale_0.5_attention_scale_0.9	5.29
ngram_lm_scale_0.5_attention_scale_1.2	5.29
ngram_lm_scale_0.5_attention_scale_1.3	5.29
ngram_lm_scale_0.5_attention_scale_1.7	5.29
ngram_lm_scale_0.6_attention_scale_1.5	5.29
ngram_lm_scale_0.3_attention_scale_1.2	5.3
ngram_lm_scale_0.3_attention_scale_1.3	5.3
ngram_lm_scale_0.3_attention_scale_1.5	5.3
ngram_lm_scale_0.5_attention_scale_1.5	5.3
ngram_lm_scale_0.5_attention_scale_1.9	5.3
ngram_lm_scale_0.1_attention_scale_1.2	5.31
ngram_lm_scale_0.1_attention_scale_1.3	5.31
ngram_lm_scale_0.1_attention_scale_1.5	5.31
ngram_lm_scale_0.1_attention_scale_1.7	5.31
ngram_lm_scale_0.3_attention_scale_0.5	5.31
ngram_lm_scale_0.5_attention_scale_0.6	5.31
ngram_lm_scale_0.5_attention_scale_0.7	5.31
ngram_lm_scale_0.6_attention_scale_1.1	5.31
ngram_lm_scale_0.6_attention_scale_1.3	5.31
ngram_lm_scale_0.7_attention_scale_2.0	5.31
ngram_lm_scale_0.7_attention_scale_2.5	5.31
ngram_lm_scale_0.01_attention_scale_1.2	5.32
ngram_lm_scale_0.05_attention_scale_1.3	5.32
ngram_lm_scale_0.05_attention_scale_1.7	5.32
ngram_lm_scale_0.08_attention_scale_1.2	5.32
ngram_lm_scale_0.08_attention_scale_1.3	5.32
ngram_lm_scale_0.08_attention_scale_1.5	5.32
ngram_lm_scale_0.08_attention_scale_1.7	5.32
ngram_lm_scale_0.1_attention_scale_1.0	5.32
ngram_lm_scale_0.1_attention_scale_1.1	5.32
ngram_lm_scale_0.3_attention_scale_1.7	5.32
ngram_lm_scale_0.5_attention_scale_2.0	5.32
ngram_lm_scale_0.5_attention_scale_2.1	5.32
ngram_lm_scale_0.5_attention_scale_2.2	5.32
ngram_lm_scale_0.5_attention_scale_2.3	5.32
ngram_lm_scale_0.5_attention_scale_2.5	5.32
ngram_lm_scale_0.6_attention_scale_0.9	5.32
ngram_lm_scale_0.6_attention_scale_1.0	5.32
ngram_lm_scale_0.6_attention_scale_1.2	5.32
ngram_lm_scale_0.6_attention_scale_1.7	5.32
ngram_lm_scale_0.6_attention_scale_1.9	5.32
ngram_lm_scale_0.6_attention_scale_2.0	5.32
ngram_lm_scale_0.6_attention_scale_2.1	5.32
ngram_lm_scale_0.6_attention_scale_2.2	5.32
ngram_lm_scale_0.6_attention_scale_2.5	5.32
ngram_lm_scale_0.7_attention_scale_1.5	5.32
ngram_lm_scale_0.7_attention_scale_1.7	5.32
ngram_lm_scale_0.7_attention_scale_1.9	5.32
ngram_lm_scale_0.7_attention_scale_2.1	5.32
ngram_lm_scale_0.7_attention_scale_2.2	5.32
ngram_lm_scale_0.7_attention_scale_2.3	5.32
ngram_lm_scale_0.01_attention_scale_1.1	5.33
ngram_lm_scale_0.01_attention_scale_1.3	5.33
ngram_lm_scale_0.01_attention_scale_1.5	5.33
ngram_lm_scale_0.01_attention_scale_1.7	5.33
ngram_lm_scale_0.05_attention_scale_1.0	5.33
ngram_lm_scale_0.05_attention_scale_1.1	5.33
ngram_lm_scale_0.05_attention_scale_1.2	5.33
ngram_lm_scale_0.05_attention_scale_1.5	5.33
ngram_lm_scale_0.05_attention_scale_1.9	5.33
ngram_lm_scale_0.08_attention_scale_0.9	5.33
ngram_lm_scale_0.08_attention_scale_1.0	5.33
ngram_lm_scale_0.08_attention_scale_1.1	5.33
ngram_lm_scale_0.08_attention_scale_1.9	5.33
ngram_lm_scale_0.08_attention_scale_2.0	5.33
ngram_lm_scale_0.08_attention_scale_2.1	5.33
ngram_lm_scale_0.1_attention_scale_0.9	5.33
ngram_lm_scale_0.1_attention_scale_1.9	5.33
ngram_lm_scale_0.1_attention_scale_2.0	5.33
ngram_lm_scale_0.1_attention_scale_2.1	5.33
ngram_lm_scale_0.3_attention_scale_1.9	5.33
ngram_lm_scale_0.5_attention_scale_0.5	5.33
ngram_lm_scale_0.6_attention_scale_2.3	5.33
ngram_lm_scale_0.01_attention_scale_1.9	5.34
ngram_lm_scale_0.01_attention_scale_2.0	5.34
ngram_lm_scale_0.01_attention_scale_2.1	5.34
ngram_lm_scale_0.01_attention_scale_2.2	5.34
ngram_lm_scale_0.01_attention_scale_2.3	5.34
ngram_lm_scale_0.05_attention_scale_0.9	5.34
ngram_lm_scale_0.05_attention_scale_2.0	5.34
ngram_lm_scale_0.05_attention_scale_2.1	5.34
ngram_lm_scale_0.05_attention_scale_2.5	5.34
ngram_lm_scale_0.08_attention_scale_2.5	5.34
ngram_lm_scale_0.1_attention_scale_2.2	5.34
ngram_lm_scale_0.1_attention_scale_2.3	5.34
ngram_lm_scale_0.1_attention_scale_2.5	5.34
ngram_lm_scale_0.3_attention_scale_2.0	5.34
ngram_lm_scale_0.3_attention_scale_2.1	5.34
ngram_lm_scale_0.3_attention_scale_2.2	5.34
ngram_lm_scale_0.3_attention_scale_2.3	5.34
ngram_lm_scale_0.3_attention_scale_2.5	5.34
ngram_lm_scale_0.5_attention_scale_3.0	5.34
ngram_lm_scale_0.6_attention_scale_0.7	5.34
ngram_lm_scale_0.6_attention_scale_3.0	5.34
ngram_lm_scale_0.7_attention_scale_3.0	5.34
ngram_lm_scale_0.01_attention_scale_2.5	5.35
ngram_lm_scale_0.01_attention_scale_3.0	5.35
ngram_lm_scale_0.05_attention_scale_2.2	5.35
ngram_lm_scale_0.05_attention_scale_2.3	5.35
ngram_lm_scale_0.08_attention_scale_2.2	5.35
ngram_lm_scale_0.08_attention_scale_2.3	5.35
ngram_lm_scale_0.1_attention_scale_0.7	5.35
ngram_lm_scale_0.1_attention_scale_3.0	5.35
ngram_lm_scale_0.3_attention_scale_3.0	5.35
ngram_lm_scale_0.6_attention_scale_4.0	5.35
ngram_lm_scale_0.7_attention_scale_1.3	5.35
ngram_lm_scale_0.7_attention_scale_4.0	5.35
ngram_lm_scale_0.01_attention_scale_0.9	5.36
ngram_lm_scale_0.01_attention_scale_1.0	5.36
ngram_lm_scale_0.05_attention_scale_3.0	5.36
ngram_lm_scale_0.08_attention_scale_0.7	5.36
ngram_lm_scale_0.08_attention_scale_3.0	5.36
ngram_lm_scale_0.3_attention_scale_0.3	5.36
ngram_lm_scale_0.3_attention_scale_4.0	5.36
ngram_lm_scale_0.5_attention_scale_4.0	5.36
ngram_lm_scale_0.6_attention_scale_0.6	5.36
ngram_lm_scale_0.7_attention_scale_1.0	5.36
ngram_lm_scale_0.7_attention_scale_1.1	5.36
ngram_lm_scale_0.01_attention_scale_4.0	5.37
ngram_lm_scale_0.05_attention_scale_0.7	5.37
ngram_lm_scale_0.05_attention_scale_4.0	5.37
ngram_lm_scale_0.08_attention_scale_4.0	5.37
ngram_lm_scale_0.1_attention_scale_0.6	5.37
ngram_lm_scale_0.1_attention_scale_4.0	5.37
ngram_lm_scale_0.5_attention_scale_5.0	5.37
ngram_lm_scale_0.6_attention_scale_0.5	5.37
ngram_lm_scale_0.6_attention_scale_5.0	5.37
ngram_lm_scale_0.7_attention_scale_0.7	5.37
ngram_lm_scale_0.7_attention_scale_0.9	5.37
ngram_lm_scale_0.7_attention_scale_1.2	5.37
ngram_lm_scale_0.7_attention_scale_5.0	5.37
ngram_lm_scale_0.9_attention_scale_2.2	5.37
ngram_lm_scale_0.9_attention_scale_2.3	5.37
ngram_lm_scale_0.9_attention_scale_3.0	5.37
ngram_lm_scale_0.05_attention_scale_0.6	5.38
ngram_lm_scale_0.08_attention_scale_0.6	5.38
ngram_lm_scale_0.08_attention_scale_5.0	5.38
ngram_lm_scale_0.1_attention_scale_0.5	5.38
ngram_lm_scale_0.1_attention_scale_5.0	5.38
ngram_lm_scale_0.9_attention_scale_2.0	5.38
ngram_lm_scale_0.9_attention_scale_2.1	5.38
ngram_lm_scale_0.9_attention_scale_2.5	5.38
ngram_lm_scale_0.01_attention_scale_0.7	5.39
ngram_lm_scale_0.01_attention_scale_5.0	5.39
ngram_lm_scale_0.05_attention_scale_5.0	5.39
ngram_lm_scale_0.08_attention_scale_0.5	5.39
ngram_lm_scale_0.3_attention_scale_5.0	5.39
ngram_lm_scale_0.5_attention_scale_0.3	5.39
ngram_lm_scale_0.9_attention_scale_1.7	5.39
ngram_lm_scale_0.9_attention_scale_1.9	5.39
ngram_lm_scale_0.9_attention_scale_4.0	5.39
ngram_lm_scale_0.9_attention_scale_5.0	5.39
ngram_lm_scale_1.0_attention_scale_2.5	5.39
ngram_lm_scale_0.01_attention_scale_0.6	5.4
ngram_lm_scale_0.05_attention_scale_0.5	5.4
ngram_lm_scale_0.7_attention_scale_0.6	5.4
ngram_lm_scale_1.0_attention_scale_2.3	5.4
ngram_lm_scale_1.0_attention_scale_3.0	5.4
ngram_lm_scale_1.0_attention_scale_4.0	5.4
ngram_lm_scale_1.0_attention_scale_5.0	5.41
ngram_lm_scale_1.1_attention_scale_4.0	5.41
ngram_lm_scale_0.6_attention_scale_0.3	5.42
ngram_lm_scale_0.9_attention_scale_1.3	5.42
ngram_lm_scale_0.9_attention_scale_1.5	5.42
ngram_lm_scale_1.0_attention_scale_2.0	5.42
ngram_lm_scale_1.0_attention_scale_2.1	5.42
ngram_lm_scale_1.0_attention_scale_2.2	5.42
ngram_lm_scale_1.1_attention_scale_3.0	5.42
ngram_lm_scale_0.01_attention_scale_0.5	5.43
ngram_lm_scale_0.1_attention_scale_0.3	5.43
ngram_lm_scale_0.9_attention_scale_1.2	5.43
ngram_lm_scale_1.0_attention_scale_1.9	5.43
ngram_lm_scale_1.1_attention_scale_5.0	5.43
ngram_lm_scale_0.08_attention_scale_0.3	5.44
ngram_lm_scale_0.3_attention_scale_0.08	5.44
ngram_lm_scale_0.3_attention_scale_0.1	5.44
ngram_lm_scale_0.7_attention_scale_0.5	5.44
ngram_lm_scale_1.0_attention_scale_1.7	5.44
ngram_lm_scale_1.1_attention_scale_2.2	5.44
ngram_lm_scale_1.1_attention_scale_2.3	5.44
ngram_lm_scale_1.1_attention_scale_2.5	5.44
ngram_lm_scale_1.2_attention_scale_4.0	5.44
ngram_lm_scale_0.9_attention_scale_1.1	5.45
ngram_lm_scale_1.0_attention_scale_1.5	5.45
ngram_lm_scale_1.1_attention_scale_2.1	5.45
ngram_lm_scale_1.2_attention_scale_5.0	5.45
ngram_lm_scale_0.05_attention_scale_0.3	5.46
ngram_lm_scale_0.3_attention_scale_0.05	5.46
ngram_lm_scale_0.5_attention_scale_0.1	5.46
ngram_lm_scale_1.2_attention_scale_3.0	5.46
ngram_lm_scale_1.3_attention_scale_5.0	5.46
ngram_lm_scale_0.5_attention_scale_0.08	5.47
ngram_lm_scale_0.9_attention_scale_1.0	5.47
ngram_lm_scale_1.1_attention_scale_2.0	5.47
ngram_lm_scale_1.2_attention_scale_2.5	5.47
ngram_lm_scale_1.3_attention_scale_4.0	5.47
ngram_lm_scale_0.01_attention_scale_0.3	5.48
ngram_lm_scale_1.0_attention_scale_1.3	5.48
ngram_lm_scale_1.1_attention_scale_1.9	5.48
ngram_lm_scale_1.2_attention_scale_2.3	5.48
ngram_lm_scale_0.3_attention_scale_0.01	5.49
ngram_lm_scale_0.5_attention_scale_0.05	5.49
ngram_lm_scale_1.1_attention_scale_1.7	5.49
ngram_lm_scale_1.2_attention_scale_2.2	5.49
ngram_lm_scale_1.3_attention_scale_3.0	5.49
ngram_lm_scale_0.9_attention_scale_0.9	5.5
ngram_lm_scale_1.0_attention_scale_1.2	5.5
ngram_lm_scale_1.5_attention_scale_5.0	5.5
ngram_lm_scale_1.2_attention_scale_2.1	5.51
ngram_lm_scale_1.5_attention_scale_4.0	5.51
ngram_lm_scale_0.1_attention_scale_0.1	5.52
ngram_lm_scale_1.2_attention_scale_2.0	5.52
ngram_lm_scale_0.08_attention_scale_0.1	5.53
ngram_lm_scale_0.1_attention_scale_0.08	5.53
ngram_lm_scale_0.5_attention_scale_0.01	5.53
ngram_lm_scale_1.2_attention_scale_1.9	5.53
ngram_lm_scale_1.3_attention_scale_2.5	5.53
ngram_lm_scale_0.08_attention_scale_0.08	5.54
ngram_lm_scale_0.1_attention_scale_0.05	5.54
ngram_lm_scale_0.7_attention_scale_0.3	5.54
ngram_lm_scale_1.0_attention_scale_1.1	5.54
ngram_lm_scale_1.1_attention_scale_1.5	5.54
ngram_lm_scale_1.3_attention_scale_2.3	5.54
ngram_lm_scale_1.7_attention_scale_5.0	5.54
ngram_lm_scale_0.05_attention_scale_0.1	5.55
ngram_lm_scale_0.05_attention_scale_0.08	5.56
ngram_lm_scale_0.08_attention_scale_0.05	5.56
ngram_lm_scale_0.6_attention_scale_0.1	5.56
ngram_lm_scale_1.2_attention_scale_1.7	5.56
ngram_lm_scale_1.3_attention_scale_2.2	5.56
ngram_lm_scale_1.5_attention_scale_3.0	5.56
ngram_lm_scale_1.0_attention_scale_1.0	5.57
ngram_lm_scale_1.1_attention_scale_1.3	5.57
ngram_lm_scale_1.3_attention_scale_2.1	5.57
ngram_lm_scale_1.7_attention_scale_4.0	5.57
ngram_lm_scale_0.1_attention_scale_0.01	5.58
ngram_lm_scale_0.6_attention_scale_0.08	5.58
ngram_lm_scale_1.9_attention_scale_5.0	5.58
ngram_lm_scale_0.01_attention_scale_0.1	5.59
ngram_lm_scale_0.05_attention_scale_0.05	5.59
ngram_lm_scale_0.6_attention_scale_0.05	5.59
ngram_lm_scale_0.9_attention_scale_0.7	5.59
ngram_lm_scale_1.3_attention_scale_2.0	5.59
ngram_lm_scale_0.01_attention_scale_0.08	5.6
ngram_lm_scale_0.08_attention_scale_0.01	5.61
ngram_lm_scale_1.1_attention_scale_1.2	5.61
ngram_lm_scale_1.2_attention_scale_1.5	5.61
ngram_lm_scale_1.3_attention_scale_1.9	5.61
ngram_lm_scale_0.01_attention_scale_0.05	5.62
ngram_lm_scale_0.6_attention_scale_0.01	5.62
ngram_lm_scale_1.0_attention_scale_0.9	5.62
ngram_lm_scale_2.0_attention_scale_5.0	5.62
ngram_lm_scale_0.05_attention_scale_0.01	5.63
ngram_lm_scale_1.5_attention_scale_2.5	5.63
ngram_lm_scale_0.9_attention_scale_0.6	5.64
ngram_lm_scale_1.1_attention_scale_1.1	5.64
ngram_lm_scale_0.01_attention_scale_0.01	5.65
ngram_lm_scale_1.3_attention_scale_1.7	5.65
ngram_lm_scale_1.9_attention_scale_4.0	5.65
ngram_lm_scale_2.1_attention_scale_5.0	5.65
ngram_lm_scale_1.1_attention_scale_1.0	5.67
ngram_lm_scale_1.2_attention_scale_1.3	5.67
ngram_lm_scale_1.5_attention_scale_2.3	5.67
ngram_lm_scale_1.7_attention_scale_3.0	5.67
ngram_lm_scale_1.5_attention_scale_2.2	5.68
ngram_lm_scale_2.2_attention_scale_5.0	5.68
ngram_lm_scale_0.7_attention_scale_0.1	5.69
ngram_lm_scale_2.0_attention_scale_4.0	5.69
ngram_lm_scale_1.0_attention_scale_0.7	5.7
ngram_lm_scale_0.9_attention_scale_0.5	5.71
ngram_lm_scale_1.2_attention_scale_1.2	5.71
ngram_lm_scale_1.3_attention_scale_1.5	5.71
ngram_lm_scale_0.7_attention_scale_0.08	5.72
ngram_lm_scale_1.1_attention_scale_0.9	5.72
ngram_lm_scale_1.5_attention_scale_2.1	5.72
ngram_lm_scale_2.3_attention_scale_5.0	5.72
ngram_lm_scale_1.2_attention_scale_1.1	5.73
ngram_lm_scale_0.7_attention_scale_0.05	5.74
ngram_lm_scale_1.5_attention_scale_2.0	5.74
ngram_lm_scale_2.1_attention_scale_4.0	5.74
ngram_lm_scale_0.7_attention_scale_0.01	5.77
ngram_lm_scale_1.0_attention_scale_0.6	5.77
ngram_lm_scale_1.7_attention_scale_2.5	5.77
ngram_lm_scale_1.3_attention_scale_1.3	5.78
ngram_lm_scale_1.5_attention_scale_1.9	5.78
ngram_lm_scale_2.2_attention_scale_4.0	5.78
ngram_lm_scale_1.2_attention_scale_1.0	5.79
ngram_lm_scale_1.9_attention_scale_3.0	5.79
ngram_lm_scale_2.5_attention_scale_5.0	5.8
ngram_lm_scale_0.9_attention_scale_0.3	5.81
ngram_lm_scale_1.3_attention_scale_1.2	5.81
ngram_lm_scale_1.0_attention_scale_0.5	5.83
ngram_lm_scale_1.2_attention_scale_0.9	5.83
ngram_lm_scale_2.3_attention_scale_4.0	5.83
ngram_lm_scale_1.5_attention_scale_1.7	5.84
ngram_lm_scale_1.7_attention_scale_2.3	5.84
ngram_lm_scale_1.1_attention_scale_0.7	5.85
ngram_lm_scale_1.3_attention_scale_1.1	5.86
ngram_lm_scale_1.7_attention_scale_2.2	5.87
ngram_lm_scale_2.0_attention_scale_3.0	5.88
ngram_lm_scale_1.5_attention_scale_1.5	5.91
ngram_lm_scale_1.7_attention_scale_2.1	5.91
ngram_lm_scale_1.1_attention_scale_0.6	5.92
ngram_lm_scale_1.3_attention_scale_1.0	5.92
ngram_lm_scale_1.7_attention_scale_2.0	5.95
ngram_lm_scale_1.9_attention_scale_2.5	5.96
ngram_lm_scale_2.1_attention_scale_3.0	5.98
ngram_lm_scale_1.0_attention_scale_0.3	5.99
ngram_lm_scale_1.2_attention_scale_0.7	5.99
ngram_lm_scale_1.7_attention_scale_1.9	5.99
ngram_lm_scale_1.1_attention_scale_0.5	6.0
ngram_lm_scale_1.3_attention_scale_0.9	6.0
ngram_lm_scale_2.5_attention_scale_4.0	6.0
ngram_lm_scale_0.9_attention_scale_0.1	6.03
ngram_lm_scale_1.5_attention_scale_1.3	6.03
ngram_lm_scale_1.9_attention_scale_2.3	6.04
ngram_lm_scale_2.2_attention_scale_3.0	6.05
ngram_lm_scale_0.9_attention_scale_0.08	6.06
ngram_lm_scale_1.2_attention_scale_0.6	6.06
ngram_lm_scale_2.0_attention_scale_2.5	6.06
ngram_lm_scale_3.0_attention_scale_5.0	6.06
ngram_lm_scale_1.9_attention_scale_2.2	6.08
ngram_lm_scale_1.5_attention_scale_1.2	6.09
ngram_lm_scale_1.7_attention_scale_1.7	6.1
ngram_lm_scale_0.9_attention_scale_0.05	6.12
ngram_lm_scale_1.9_attention_scale_2.1	6.13
ngram_lm_scale_2.0_attention_scale_2.3	6.14
ngram_lm_scale_2.3_attention_scale_3.0	6.15
ngram_lm_scale_2.1_attention_scale_2.5	6.16
ngram_lm_scale_0.9_attention_scale_0.01	6.17
ngram_lm_scale_1.3_attention_scale_0.7	6.17
ngram_lm_scale_1.5_attention_scale_1.1	6.17
ngram_lm_scale_1.9_attention_scale_2.0	6.17
ngram_lm_scale_1.2_attention_scale_0.5	6.19
ngram_lm_scale_1.7_attention_scale_1.5	6.19
ngram_lm_scale_2.0_attention_scale_2.2	6.19
ngram_lm_scale_1.1_attention_scale_0.3	6.21
ngram_lm_scale_1.9_attention_scale_1.9	6.22
ngram_lm_scale_2.0_attention_scale_2.1	6.24
ngram_lm_scale_2.1_attention_scale_2.3	6.25
ngram_lm_scale_1.5_attention_scale_1.0	6.27
ngram_lm_scale_2.2_attention_scale_2.5	6.27
ngram_lm_scale_1.3_attention_scale_0.6	6.28
ngram_lm_scale_1.0_attention_scale_0.1	6.29
ngram_lm_scale_2.0_attention_scale_2.0	6.29
ngram_lm_scale_2.1_attention_scale_2.2	6.3
ngram_lm_scale_2.5_attention_scale_3.0	6.32
ngram_lm_scale_1.0_attention_scale_0.08	6.33
ngram_lm_scale_3.0_attention_scale_4.0	6.34
ngram_lm_scale_2.1_attention_scale_2.1	6.35
ngram_lm_scale_1.7_attention_scale_1.3	6.36
ngram_lm_scale_2.2_attention_scale_2.3	6.36
ngram_lm_scale_1.5_attention_scale_0.9	6.37
ngram_lm_scale_1.9_attention_scale_1.7	6.37
ngram_lm_scale_2.0_attention_scale_1.9	6.37
ngram_lm_scale_2.3_attention_scale_2.5	6.37
ngram_lm_scale_1.0_attention_scale_0.05	6.39
ngram_lm_scale_2.2_attention_scale_2.2	6.43
ngram_lm_scale_1.3_attention_scale_0.5	6.44
ngram_lm_scale_2.1_attention_scale_2.0	6.44
ngram_lm_scale_1.7_attention_scale_1.2	6.46
ngram_lm_scale_1.0_attention_scale_0.01	6.49
ngram_lm_scale_1.2_attention_scale_0.3	6.49
ngram_lm_scale_2.1_attention_scale_1.9	6.5
ngram_lm_scale_2.3_attention_scale_2.3	6.5
ngram_lm_scale_2.0_attention_scale_1.7	6.51
ngram_lm_scale_2.2_attention_scale_2.1	6.51
ngram_lm_scale_1.9_attention_scale_1.5	6.54
ngram_lm_scale_2.3_attention_scale_2.2	6.56
ngram_lm_scale_2.2_attention_scale_2.0	6.57
ngram_lm_scale_1.1_attention_scale_0.1	6.58
ngram_lm_scale_1.7_attention_scale_1.1	6.58
ngram_lm_scale_2.5_attention_scale_2.5	6.62
ngram_lm_scale_1.1_attention_scale_0.08	6.63
ngram_lm_scale_1.5_attention_scale_0.7	6.65
ngram_lm_scale_2.3_attention_scale_2.1	6.65
ngram_lm_scale_2.2_attention_scale_1.9	6.68
ngram_lm_scale_1.7_attention_scale_1.0	6.7
ngram_lm_scale_2.1_attention_scale_1.7	6.7
ngram_lm_scale_1.1_attention_scale_0.05	6.72
ngram_lm_scale_2.0_attention_scale_1.5	6.72
ngram_lm_scale_2.3_attention_scale_2.0	6.74
ngram_lm_scale_4.0_attention_scale_5.0	6.75
ngram_lm_scale_1.9_attention_scale_1.3	6.76
ngram_lm_scale_2.5_attention_scale_2.3	6.77
ngram_lm_scale_1.1_attention_scale_0.01	6.81
ngram_lm_scale_1.3_attention_scale_0.3	6.81
ngram_lm_scale_1.5_attention_scale_0.6	6.82
ngram_lm_scale_2.3_attention_scale_1.9	6.82
ngram_lm_scale_1.7_attention_scale_0.9	6.83
ngram_lm_scale_2.2_attention_scale_1.7	6.85
ngram_lm_scale_1.9_attention_scale_1.2	6.87
ngram_lm_scale_2.5_attention_scale_2.2	6.87
ngram_lm_scale_2.1_attention_scale_1.5	6.89
ngram_lm_scale_3.0_attention_scale_3.0	6.9
ngram_lm_scale_2.0_attention_scale_1.3	6.94
ngram_lm_scale_1.2_attention_scale_0.1	6.95
ngram_lm_scale_2.5_attention_scale_2.1	6.95
ngram_lm_scale_1.9_attention_scale_1.1	6.98
ngram_lm_scale_1.2_attention_scale_0.08	7.0
ngram_lm_scale_1.5_attention_scale_0.5	7.0
ngram_lm_scale_2.3_attention_scale_1.7	7.05
ngram_lm_scale_2.5_attention_scale_2.0	7.05
ngram_lm_scale_2.0_attention_scale_1.2	7.06
ngram_lm_scale_1.2_attention_scale_0.05	7.09
ngram_lm_scale_2.2_attention_scale_1.5	7.09
ngram_lm_scale_2.1_attention_scale_1.3	7.14
ngram_lm_scale_2.5_attention_scale_1.9	7.14
ngram_lm_scale_1.9_attention_scale_1.0	7.15
ngram_lm_scale_1.7_attention_scale_0.7	7.2
ngram_lm_scale_1.2_attention_scale_0.01	7.22
ngram_lm_scale_2.0_attention_scale_1.1	7.23
ngram_lm_scale_2.1_attention_scale_1.2	7.28
ngram_lm_scale_2.3_attention_scale_1.5	7.28
ngram_lm_scale_3.0_attention_scale_2.5	7.28
ngram_lm_scale_4.0_attention_scale_4.0	7.29
ngram_lm_scale_1.3_attention_scale_0.1	7.33
ngram_lm_scale_1.9_attention_scale_0.9	7.33
ngram_lm_scale_2.2_attention_scale_1.3	7.36
ngram_lm_scale_1.7_attention_scale_0.6	7.38
ngram_lm_scale_2.0_attention_scale_1.0	7.38
ngram_lm_scale_2.5_attention_scale_1.7	7.39
ngram_lm_scale_1.3_attention_scale_0.08	7.4
ngram_lm_scale_2.1_attention_scale_1.1	7.47
ngram_lm_scale_1.5_attention_scale_0.3	7.48
ngram_lm_scale_1.3_attention_scale_0.05	7.51
ngram_lm_scale_3.0_attention_scale_2.3	7.52
ngram_lm_scale_2.2_attention_scale_1.2	7.54
ngram_lm_scale_2.3_attention_scale_1.3	7.6
ngram_lm_scale_5.0_attention_scale_5.0	7.6
ngram_lm_scale_2.0_attention_scale_0.9	7.61
ngram_lm_scale_1.7_attention_scale_0.5	7.63
ngram_lm_scale_3.0_attention_scale_2.2	7.64
ngram_lm_scale_1.3_attention_scale_0.01	7.66
ngram_lm_scale_2.1_attention_scale_1.0	7.68
ngram_lm_scale_2.5_attention_scale_1.5	7.7
ngram_lm_scale_2.2_attention_scale_1.1	7.73
ngram_lm_scale_1.9_attention_scale_0.7	7.75
ngram_lm_scale_2.3_attention_scale_1.2	7.77
ngram_lm_scale_3.0_attention_scale_2.1	7.77
ngram_lm_scale_2.1_attention_scale_0.9	7.89
ngram_lm_scale_3.0_attention_scale_2.0	7.92
ngram_lm_scale_2.2_attention_scale_1.0	7.94
ngram_lm_scale_2.3_attention_scale_1.1	7.99
ngram_lm_scale_1.9_attention_scale_0.6	8.05
ngram_lm_scale_2.5_attention_scale_1.3	8.07
ngram_lm_scale_3.0_attention_scale_1.9	8.09
ngram_lm_scale_2.0_attention_scale_0.7	8.1
ngram_lm_scale_1.5_attention_scale_0.1	8.17
ngram_lm_scale_2.2_attention_scale_0.9	8.19
ngram_lm_scale_4.0_attention_scale_3.0	8.2
ngram_lm_scale_2.3_attention_scale_1.0	8.23
ngram_lm_scale_1.5_attention_scale_0.08	8.26
ngram_lm_scale_1.7_attention_scale_0.3	8.27
ngram_lm_scale_2.5_attention_scale_1.2	8.27
ngram_lm_scale_1.9_attention_scale_0.5	8.35
ngram_lm_scale_2.0_attention_scale_0.6	8.37
ngram_lm_scale_3.0_attention_scale_1.7	8.39
ngram_lm_scale_5.0_attention_scale_4.0	8.39
ngram_lm_scale_2.1_attention_scale_0.7	8.4
ngram_lm_scale_1.5_attention_scale_0.05	8.41
ngram_lm_scale_2.3_attention_scale_0.9	8.46
ngram_lm_scale_2.5_attention_scale_1.1	8.49
ngram_lm_scale_1.5_attention_scale_0.01	8.56
ngram_lm_scale_2.0_attention_scale_0.5	8.67
ngram_lm_scale_2.1_attention_scale_0.6	8.68
ngram_lm_scale_2.2_attention_scale_0.7	8.7
ngram_lm_scale_2.5_attention_scale_1.0	8.74
ngram_lm_scale_3.0_attention_scale_1.5	8.78
ngram_lm_scale_4.0_attention_scale_2.5	8.85
ngram_lm_scale_2.5_attention_scale_0.9	8.97
ngram_lm_scale_2.3_attention_scale_0.7	8.98
ngram_lm_scale_2.1_attention_scale_0.5	8.99
ngram_lm_scale_2.2_attention_scale_0.6	8.99
ngram_lm_scale_1.9_attention_scale_0.3	9.01
ngram_lm_scale_1.7_attention_scale_0.1	9.04
ngram_lm_scale_1.7_attention_scale_0.08	9.13
ngram_lm_scale_4.0_attention_scale_2.3	9.13
ngram_lm_scale_3.0_attention_scale_1.3	9.15
ngram_lm_scale_4.0_attention_scale_2.2	9.25
ngram_lm_scale_1.7_attention_scale_0.05	9.26
ngram_lm_scale_2.3_attention_scale_0.6	9.26
ngram_lm_scale_2.2_attention_scale_0.5	9.29
ngram_lm_scale_2.0_attention_scale_0.3	9.35
ngram_lm_scale_3.0_attention_scale_1.2	9.37
ngram_lm_scale_4.0_attention_scale_2.1	9.4
ngram_lm_scale_5.0_attention_scale_3.0	9.4
ngram_lm_scale_1.7_attention_scale_0.01	9.44
ngram_lm_scale_2.5_attention_scale_0.7	9.49
ngram_lm_scale_2.3_attention_scale_0.5	9.55
ngram_lm_scale_4.0_attention_scale_2.0	9.56
ngram_lm_scale_3.0_attention_scale_1.1	9.59
ngram_lm_scale_2.1_attention_scale_0.3	9.64
ngram_lm_scale_4.0_attention_scale_1.9	9.68
ngram_lm_scale_1.9_attention_scale_0.1	9.75
ngram_lm_scale_2.5_attention_scale_0.6	9.75
ngram_lm_scale_3.0_attention_scale_1.0	9.8
ngram_lm_scale_1.9_attention_scale_0.08	9.83
ngram_lm_scale_2.2_attention_scale_0.3	9.9
ngram_lm_scale_1.9_attention_scale_0.05	9.92
ngram_lm_scale_5.0_attention_scale_2.5	9.95
ngram_lm_scale_4.0_attention_scale_1.7	9.98
ngram_lm_scale_3.0_attention_scale_0.9	10.0
ngram_lm_scale_2.0_attention_scale_0.1	10.01
ngram_lm_scale_2.5_attention_scale_0.5	10.02
ngram_lm_scale_1.9_attention_scale_0.01	10.05
ngram_lm_scale_2.0_attention_scale_0.08	10.07
ngram_lm_scale_2.3_attention_scale_0.3	10.11
ngram_lm_scale_2.0_attention_scale_0.05	10.16
ngram_lm_scale_5.0_attention_scale_2.3	10.19
ngram_lm_scale_2.1_attention_scale_0.1	10.24
ngram_lm_scale_4.0_attention_scale_1.5	10.25
ngram_lm_scale_2.0_attention_scale_0.01	10.3
ngram_lm_scale_5.0_attention_scale_2.2	10.3
ngram_lm_scale_2.1_attention_scale_0.08	10.31
ngram_lm_scale_3.0_attention_scale_0.7	10.38
ngram_lm_scale_5.0_attention_scale_2.1	10.39
ngram_lm_scale_2.1_attention_scale_0.05	10.4
ngram_lm_scale_2.2_attention_scale_0.1	10.45
ngram_lm_scale_2.5_attention_scale_0.3	10.48
ngram_lm_scale_2.1_attention_scale_0.01	10.5
ngram_lm_scale_2.2_attention_scale_0.08	10.5
ngram_lm_scale_5.0_attention_scale_2.0	10.5
ngram_lm_scale_4.0_attention_scale_1.3	10.52
ngram_lm_scale_3.0_attention_scale_0.6	10.55
ngram_lm_scale_2.2_attention_scale_0.05	10.58
ngram_lm_scale_5.0_attention_scale_1.9	10.6
ngram_lm_scale_2.3_attention_scale_0.1	10.63
ngram_lm_scale_4.0_attention_scale_1.2	10.64
ngram_lm_scale_2.3_attention_scale_0.08	10.69
ngram_lm_scale_2.2_attention_scale_0.01	10.7
ngram_lm_scale_3.0_attention_scale_0.5	10.75
ngram_lm_scale_2.3_attention_scale_0.05	10.77
ngram_lm_scale_4.0_attention_scale_1.1	10.78
ngram_lm_scale_5.0_attention_scale_1.7	10.8
ngram_lm_scale_2.3_attention_scale_0.01	10.86
ngram_lm_scale_4.0_attention_scale_1.0	10.91
ngram_lm_scale_2.5_attention_scale_0.1	10.93
ngram_lm_scale_2.5_attention_scale_0.08	10.98
ngram_lm_scale_5.0_attention_scale_1.5	11.01
ngram_lm_scale_4.0_attention_scale_0.9	11.03
ngram_lm_scale_2.5_attention_scale_0.05	11.05
ngram_lm_scale_3.0_attention_scale_0.3	11.1
ngram_lm_scale_2.5_attention_scale_0.01	11.13
ngram_lm_scale_5.0_attention_scale_1.3	11.2
ngram_lm_scale_4.0_attention_scale_0.7	11.28
ngram_lm_scale_5.0_attention_scale_1.2	11.29
ngram_lm_scale_5.0_attention_scale_1.1	11.39
ngram_lm_scale_4.0_attention_scale_0.6	11.41
ngram_lm_scale_3.0_attention_scale_0.1	11.44
ngram_lm_scale_3.0_attention_scale_0.08	11.48
ngram_lm_scale_5.0_attention_scale_1.0	11.48
ngram_lm_scale_3.0_attention_scale_0.05	11.52
ngram_lm_scale_4.0_attention_scale_0.5	11.52
ngram_lm_scale_3.0_attention_scale_0.01	11.56
ngram_lm_scale_5.0_attention_scale_0.9	11.57
ngram_lm_scale_5.0_attention_scale_0.7	11.72
ngram_lm_scale_4.0_attention_scale_0.3	11.73
ngram_lm_scale_5.0_attention_scale_0.6	11.8
ngram_lm_scale_4.0_attention_scale_0.1	11.88
ngram_lm_scale_5.0_attention_scale_0.5	11.88
ngram_lm_scale_4.0_attention_scale_0.08	11.9
ngram_lm_scale_4.0_attention_scale_0.05	11.92
ngram_lm_scale_4.0_attention_scale_0.01	11.95
ngram_lm_scale_5.0_attention_scale_0.3	11.98
ngram_lm_scale_5.0_attention_scale_0.1	12.11
ngram_lm_scale_5.0_attention_scale_0.08	12.13
ngram_lm_scale_5.0_attention_scale_0.05	12.14
ngram_lm_scale_5.0_attention_scale_0.01	12.16

2024-08-18 12:10:24,931 INFO [decode.py:565] Done!
### 安装和配置新一代 Kaldi (sherpa-onnx) on 树莓派 #### 准备工作 确保树莓派已更新至最新版本的操作系统并安装必要的依赖项。对于基于 Debian 或 Ubuntu 的发行版,可以使用以下命令来更新软件包列表并升级现有软件包: ```bash sudo apt update && sudo apt upgrade -y ``` #### 安装依赖库 为了使 `sherpa-onnx` 正常运行,在树莓派上还需要额外安装一些Python库和其他工具链。执行如下指令完成这些前置条件的部署[^2]: ```bash sudo apt install python3-pip libasound2-dev portaudio19-dev pip3 install --upgrade pip setuptools wheel ``` #### 获取预编译模型文件 下载适用于 ARM 架构的 ONNX 模型以及相应的解码资源。可以从官方GitHub仓库或者其他可信源获取所需资源,并将其放置于合适的位置以便后续调用。 ```bash mkdir -p ~/kaldi-models && cd ~/kaldi-models wget https://github.com/k2-fsa/sherpa/releases/download/v0.6.0/offline-zip-ctc-aishell.zip unzip offline-zip-ctc-aishell.zip ``` #### 安装 Sherpa-Onnx Python 包 通过 PyPI 来安装最新的 `sherpa-onnx` 版本,这一步骤会自动处理大部分兼容性和路径设置问题。 ```bash pip3 install sherpa-onnx ``` #### 编写测试脚本 创建一个新的 Python 文件用于验证安装是否成功,同时也可以作为未来开发的基础框架。 ```python import wave from pathlib import Path import sherpa_onnx def main(): recognizer = sherpa_onnx.Recognizer( tokens=Path("~/kaldi-models/tokens.txt").expanduser(), encoder=Path("~/kaldi-models/encoder.onnx").expanduser(), decoder=Path("~/kaldi-models/decoder.onnx").expanduser(), joiner=Path("~/kaldi-models/joiner.onnx").expanduser() ) with wave.open("/path/to/test.wav") as f: samples = f.readframes(f.getnframes()) result = recognizer.decode(samples) print(result) if __name__ == "__main__": main() ``` 以上就是关于如何在树莓派设备上安装和初步应用新一代 Kaldi (`sherpa-onnx`) 进行语音识别的过程介绍。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

@李思成

你的鼓励将是我创作的最大动力

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付:¥1
获取中
扫码支付

您的余额不足,请更换扫码支付或充值

打赏作者

实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值