mxnet的example下有speech_recognition，基于百度的?DeepSpeech2，嘗試用該工程進行語音識別的訓練

https://github.com/apache/incubator-mxnet/tree/master/example/speech_recognition

該工程需要自己編譯帶 WarpCTC 的 mxnet 版本

我首先使用XUbuntu19.10操作系統，在編譯warp-ctc時遇到如下報錯：

/usr/local/cuda/include/crt/host_config.h:138:2: error: #error -- unsupported GNU version! gcc versions later than 8 are not supported!
我這個電腦上有GPU，報錯說不支持版本高于8的gcc，XUbuntu的19.10的gcc版本為9.2.1

然后，我就只能在XUbuntu18.04版本上進行編譯，在編譯mxnet時報如下錯誤：

CMake 3.13 or higher is required. ?You are running version 3.10.2
又嫌棄cmake版本太低，我再次換操作系統

操作系統XUbuntu19.10的cmake版本為3.13.4，gcc版本為8.3.0，看來符合上面苛刻的條件

編譯安裝warp-ctc工程

git clone https://github.com/baidu-research/warp-ctc.git
cd warp-ctc
mkdir build
cd build
cmake ../
make
sudo make install

編譯mxnet

https://mxnet.apache.org/get_started/ubuntu_setup

sudo apt install ninja-build ccache libopenblas-dev libopencv-dev
git clone --recursive https://github.com/apache/incubator-mxnet.git mxnet
cd mxnet

修改文件make/config.mk，取消注釋下面兩行

WARPCTC_PATH = $(HOME)/warp-ctc
MXNET_PLUGINS += plugin/warpctc/warpctc.mk

如果要添加CUDA支持，設置

USE_CUDA = 1
USE_CUDNN = 1

開始編譯

make -j8

支持GPU版本，參考命令

make -j8 USE_OPENCV=1 USE_BLAS=openblas USE_CUDA=1 USE_CUDA_PATH=/usr/local/cuda USE_CUDNN=1

安裝到系統中

cd python
pip install --user -e .

安裝配套

pip install mxboard
pip install soundfile

先下載兩個音頻文件，訓練試一下：

https://github.com/samsungsds-rnd/deepspeech.mxnet/tree/master/Libri_sample

把上面鏈接整個目錄都下載到目錄example/speech_recognition/下，并新建文件Libri_sample.json，內容如下：

{"duration": 2.9450625, "text": "and sharing her house which was near by", "key": "./Libri_sample/3830-12531-0030.wav"}
{"duration": 3.94, "text": "we were able to impart the information that we wanted", "key": "./Libri_sample/3830-12529-0005.wav"}

執行如下命令：

mkdir checkpoints
mkdir log
python main.py --configfile default.cfg

如果報錯：libwarpctc.so: cannot open shared object file: No such file or directory

則執行命令 ：

export LD_LIBRARY_PATH=/usr/local/lib

如果上面編譯mxnet的時候，沒有開啟支持GPU，會報錯MXNetError: Compile with USE_CUDA=1 to enable GPU usage，

如果是想在CPU上進行訓練，需要修改default.cfg，

context = cpu0

訓練完成后就可以預測了，復制default.cfg文件為predict.cfg，把

mode = train
改為

mode = predict
然后執行命令:

python main.py --configfile predict.cfg

我重新在我的神船筆記本(帶RTX2060)上訓練，預測結果為：

[    INFO][2020/01/13 16:54:22.581] label: we were able to impart the information that we wanted
[    INFO][2020/01/13 16:54:22.581] pred : we were able to impart the information that we wanted , cer: 0.000000 (distance: 0/ label length: 53)
[    INFO][2020/01/13 16:54:22.582] label: and sharing her house which was near by
[    INFO][2020/01/13 16:54:22.582] pred : and sharing her house which was near by , cer: 0.000000 (distance: 0/ label length: 39)

完全正確，好激動，真的可以預測

?

下面開始玩大的，多下載一些訓練數據，準備好250G的磁盤空間：

http://www.openslr.org/resources/12/dev-clean.tar.gz
#http://www.openslr.org/resources/12/dev-other.tar.gz
http://www.openslr.org/resources/12/test-clean.tar.gz
#http://www.openslr.org/resources/12/test-other.tar.gz
http://www.openslr.org/resources/12/train-clean-100.tar.gz
#http://www.openslr.org/resources/12/train-clean-360.tar.gz
#http://www.openslr.org/resources/12/train-other-500.tar.gz

然后全部解壓縮，

tar xvf dev-clean.tar.gz
#tar xvf dev-other.tar.gz
tar xvf test-clean.tar.gz
#tar xvf test-other.tar.gz
tar xvf train-clean-100.tar.gz
#tar xvf train-clean-360.tar.gz
#tar xvf train-other-500.tar.gz

然后轉碼為wav，執行example/speech_recognition/flac_to_wav.sh

./flac_to_wav.sh

這個命令會把子目錄下的所有flac文件轉碼為wav，保存到同目錄下

然后下載工程?https://github.com/baidu-research/ba-dls-deepspeech.git?，沒辦法，還是要用到該工程下的create_desc_json.py文件：

python create_desc_json.py ~/LibriSpeech/train-clean-100 train_corpus.json
python create_desc_json.py ~/LibriSpeech/dev-clean validation_corpus.json
python create_desc_json.py ~/LibriSpeech/test-clean test_corpus.json

然后稍微修改一下deepspeech.cfg，

train_json = train_corpus.json
test_json = test_corpus.json
val_json = validation_corpus.json

開始訓練

mkdir checkpoints
mkdir log
python main.py --configfile deepspeech.cfg

我測試感覺這個訓練極耗內存，我神船的16G ddr4根本扛不住，運行起來就會報“已殺死”，所以只好想辦法增加了64G的虛擬內存

sudo dd if=/dev/zero of=swapfile bs=64M count=1024
sudo mkswap swapfile
sudo swapon swapfile

看起來差不多虛擬內存也要耗費46G左右，加上本身的16G內存，估計共需要64G以上的內存才能運行

把該配置文件按照上面的方法，修改為predict.cfg，可以嘗試下語音識別的效果

總結

以上是生活随笔為你收集整理的mxnet speech_recognition踩坑记的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。