wav2letter++ tutorial受挫

screen wav2letter/build/Train train --flagsfile wav2letter/tutorials/1-librispeech_clean/train.cfg

Step 1: Data Preparation

For a speech recognition dataset, we usually have access to a set of audio files and their transcriptions. Create an experiment path and download the dataset.

>W2LDIR=/home/$USER/w2l        

>mkdir -p$W2LDIR

>wget -qO- http://www.openslr.org/resources/12/train-clean-100.tar.gz | tar xvz -C$W2LDIR

>wget -qO- http://www.openslr.org/resources/12/dev-clean.tar.gz | tar xvz -C$W2LDIR

>wget -qO- http://www.openslr.org/resources/12/test-clean.tar.gz | tar xvz -C$W2LDIR

此处文件不容易下载，实在不行就用下载工具然后自己解压。

wav2letter/tutorials/1-librispeech_clean/prepare_data.py --src $W2LDIR/LibriSpeech/ --dst $W2LDIR

wav2letter/tutorials/1-librispeech_clean/prepare_lm.py --dst $W2LDIR

prepare_lm.py 要下载一个3-gram.arpa.gz，又是超慢。忍受不了一样下载工具上，然后修改prepare_lm.py , 修改如下：

os.system(

        "gunzip {lm}.arpa.gz -c > {o}".format(lm=lm, o=arpa_file)

    )

Step 2: Training the Acoustic Model

首先将wav2letter/tutorials/1-librispeech_clean/train.cfg文件中的[...]替换成正确的路径

--datadir=/home/xxx/w2l/

--tokensdir=/home/xxx/w2l/

--rundir=/home/xxx/w2l/saved_models

--archdir=/home/xxx/wav2letter/tutorials/1-librispeech_clean/

然后执行：

/home/xxx/wav2letter/build/Train train --flagsfile /home/xxx/wav2letter/tutorials/1-librispeech_clean/train.cfg

训练后的logs保存在你设置的rundir目录下。

然后就code dump，据说libsnd有问题，有空在查。

terminate called after throwing an instance of 'terminate called recursively

terminate called recursively

terminate called recursively

*** Aborted at 1565019127 (unix time) try "date -d @1565019127" if you are using GNU date ***

terminate called recursively

terminate called recursively

std::runtime_error'

  what():  loadSoundInfo: unknown format or could not open stream

terminate called recursively

terminate called recursively

terminate called recursively

terminate called recursively

terminate called recursively

terminate called recursively

PC: @    0x7f8ae13dd428 gsignal

*** SIGABRT (@0x3e8000007d1) received by PID 2001 (TID 0x7f8ab1fff700) from PID 2001; stack trace: ***

    @    0x7f8ae20c7390 (unknown)

    @    0x7f8ae13dd428 gsignal

    @    0x7f8ae13df02a abort

    @    0x7f8b0033906d __gnu_cxx::__verbose_terminate_handler()

    @    0x7f8b002aa436 __cxxabiv1::__terminate()

    @    0x7f8b0032e349 __cxa_call_terminate

    @    0x7f8b002a9088 __gxx_personality_v0

    @    0x7f8b347c9aab __trunctfsf2

    @    0x7f8b347c9f49 __trunctfdf2

    @          0x5dc36b w2l::loadSoundInfo()

    @          0x5d22ec _ZN3w2l23W2lNumberedFilesDataset15loadSampleSizesEv._omp_fn.0

    @    0x7f8b347e56d5 gomp_display_affinity_thread

    @    0x7f8ae20bd6ba start_thread

    @    0x7f8ae14af41d clone

    @                0x0 (unknown)

Aborted (core dumped)