Python使用mfcc的两种方式详解

Python使用mfcc的两种方式详解，具有很好的参考价值。文章源自菜鸟学院-https://www.cainiaoxueyuan.com/bc/9227.html

1、Librosa文章源自菜鸟学院-https://www.cainiaoxueyuan.com/bc/9227.html

import librosa 
filepath = "/Users/birenjianmo/Desktop/learn/librosa/mp3/in.wav"
y,sr = librosa.load(filepath)
mfcc = librosa.feature.mfcc( y,sr,n_mfcc=13 )

返回结构为(13,None)的np.Array，None表示任意数量文章源自菜鸟学院-https://www.cainiaoxueyuan.com/bc/9227.html

2、python_speech_features文章源自菜鸟学院-https://www.cainiaoxueyuan.com/bc/9227.html

from python_speech_features import mfcc as pmfcc
filepath = "/Users/birenjianmo/Desktop/learn/librosa/mp3/in.wav"
(rate,sig) = wav.read(filepath)
amfcc = pmfcc( sig, rate ).T

返回结构为(None,13)的np.Array，通过 .T 可转为(13,None)文章源自菜鸟学院-https://www.cainiaoxueyuan.com/bc/9227.html

3、这两种方式的mfcc还是有明显的区别的，上面两个子图是从（1）Librosa得到的 mfcc[0] 和 mfcc[1]，下面的是（2）python_speech_features得到的 amfcc[0] 和 amfcc[1]文章源自菜鸟学院-https://www.cainiaoxueyuan.com/bc/9227.html

文章源自菜鸟学院-https://www.cainiaoxueyuan.com/bc/9227.html