MFCC to audio, the effect should not be too funny >V 2022-07-30 10:14:00 【Begonia_cat】 MFCCA good representation of the frequency characteristics of the audio.MFCCCan represent audio,as input to a deep learning end-to-end model（Although experimentally proven,Spectrogram ratioMFCC有更好的性能）.那么MFCCWhat is the effect of converting audio？This article will be based on Bach《小步舞曲》为例,进行实验.Oh！I don't know how to upload the audio! So this article I will just show the Experimental process. If you know how to do that, please tell me. And then I will supplement this article.（I have recently had a feverish interest in English,So when writing articles, I want to practice English）提取《小步舞曲》的MFCCimport librosa.display import matplotlib.pyplot as plt import numpy as np music = "D:/my life/music/bach.mp3" y, sr = librosa.load(music, sr=22050) n_fft = 2048 hop_length = 512 # MFCC MFCCs = librosa.feature.mfcc(y, n_fft=n_fft, hop_length=hop_length, n_mfcc=13) librosa.display.specshow(MFCCs, sr=sr, hop_length=hop_length) # specshow()Used to show graphs with a horizontal axis to the right of a vertical axis plt.xlabel("Time") plt.ylabel("MFCC") plt.colorbar() plt.show() 将MFCC转音频y= librosa.feature.inverse.mfcc_to_audio(MFCCs) # 将MFCC写入本地 import soundfile soundfile.write("D:/my life/music/bach.wav", y, 22050) # 注意：只能写成WAV格式的 如果使用的是jupyter notebook进行编程,Audio listening can be performed directly using the code below# Hear how it works import IPython.display as ipd ipd.Audio("D:/my life/music/bach.wav") From the listening effect,节奏、The strength of the notes restores the original audio very well,But the aural effects turned into percussion,即：Pitch information is lost.实验结论MFCCThe reconstruction of the converted audio is not perfect.Because the original audio is transferred to MFCC时,会丢失大量的信息.

MFCCA good representation of the frequency characteristics of the audio.MFCCCan represent audio,as input to a deep learning end-to-end model（Although experimentally proven,Spectrogram ratioMFCC有更好的性能）.那么MFCCWhat is the effect of converting audio？This article will be based on Bach《小步舞曲》为例,进行实验.

Oh！I don't know how to upload the audio! So this article I will just show the Experimental process. If you know how to do that, please tell me. And then I will supplement this article.（I have recently had a feverish interest in English,So when writing articles, I want to practice English）

提取《小步舞曲》的MFCC

import librosa.display
import matplotlib.pyplot as plt
import numpy as np

music = "D:/my life/music/bach.mp3"
y, sr = librosa.load(music, sr=22050)

n_fft = 2048
hop_length = 512

# MFCC
MFCCs = librosa.feature.mfcc(y, n_fft=n_fft, hop_length=hop_length, n_mfcc=13)
librosa.display.specshow(MFCCs, sr=sr, hop_length=hop_length)  # specshow()Used to show graphs with a horizontal axis to the right of a vertical axis
plt.xlabel("Time")
plt.ylabel("MFCC")
plt.colorbar()
plt.show()

将MFCC转音频

y= librosa.feature.inverse.mfcc_to_audio(MFCCs)

# 将MFCC写入本地
import soundfile
soundfile.write("D:/my life/music/bach.wav", y, 22050)  # 注意：只能写成WAV格式的

如果使用的是jupyter notebook进行编程,Audio listening can be performed directly using the code below

# Hear how it works
import IPython.display as ipd
ipd.Audio("D:/my life/music/bach.wav")

From the listening effect,节奏、The strength of the notes restores the original audio very well,But the aural effects turned into percussion,即：Pitch information is lost.

实验结论

MFCCThe reconstruction of the converted audio is not perfect.Because the original audio is transferred to MFCC时,会丢失大量的信息.