Kaldi 提取MFCC40维的参数设置

1. 数据准备：wave文件，，获取wav.scp，spk2utt，utt2spk三个文件

find /*/16kwav -name '*.wav' | awk -F '/' '{print $NF " " $0}' > ./data/wav.scp

find /*/16kwav -name '*.wav' | awk -F '/' '{print $NF " " $NF}' > ./data/spk2utt

find /*/16kwav -name '*.wav' | awk -F '/' '{print $NF " " $NF}' > ./data/utt2spk

2. 特征提取

首先需要更改conf/mfcc.conf文件参数，更改如下：

# config for high-resolution MFCC features, intended for neural network training. # Note: we keep all cepstra, so it has the same info as filterbank features, # but MFCC is more easily compressible (because less correlated) which is why # we prefer this method. --use-energy=false       # use average of log energy, not energy. --sample-frequency=16000 # AISHELL-2 is sampled at 16kHz --num-mel-bins=40        # similar to Google's setup. --num-ceps=40            # there is no dimensionality reduction. --low-freq=20            # low cutoff frequency for mel bins --high-freq=-400         # high cutoff frequency, relative to Nyquist of 8000 (=7600)

接下来运行如下命令：

utils/fix_data_dir.sh       /*/data

./steps/make_mfcc.sh  /*/data  ./ts_log /*/data/mfcc