Meyda 特征提取的数据在命令行和通过 nodejs 不同
[英]Meyda feature extracted dat is different in command line and through nodejs
问题描述
I tried to extract feature of an audio file using Meyda.
我尝试使用 Meyda 提取音频文件的特征。 But feature exytracted values are different when extracted using command line meyda and through javascript file但是当使用命令行 meyda 和通过 javascript 文件提取时，特征提取值是不同的
Command line
命令行meyda 1600153162.63571.wav mfcc
梅达 1600153162.63571.wav mfcc
Result结果
0.057628476337413304,0.025103534166687494,-0.0033402018588145435,-0.0050914619032203364,-0.005969043107840086,-0.009452339628697326,-0.0077781823184730455,-0.0024033016732702066,-0.0033368248853514206,-0.0012372591871309979,0.00003587458387239886,0.00293735474906295,0.0033156231532707036 1.4306079393227265,1.3713938566968262,1.2306487569743767,1.0515807610571597,0.8414768385049735,0.6060415036630535,0.37326507523088365,0.1654296077929904,-0.015392187854876595,-0.1652509593657213,-0.27571446879975764,-0.340564139593766,-0.36505384389267703 9.282664122642018,6.880236029290547,3.3870794551074606,1.7170683624244123,-0.29657914336587826,-3.18544774959499,-4.433789386396694,-3.7101490665746217,-2.816738558429225,-2.2736417577176273,-1.6636438686466657,-0.8537746809738949,-0.5116463089021152 25.65632094501052,22.93621824737729,16.08611763344924,7.490656903754756,-0.4358888007314062,-5.641656791337735,-7.229285832273481,-5.635856022395436,-2.1681330598896817,1.5575050036371694,4.29638119563214
Through javscript file
通过 javascript 文件var express = require('express'); var app = express(); const fs = require('fs'); var Meyda = require('meyda'); var load = require('audio-loader') let filename = "1600153162.63571.wav" load(filename).then(function (buffer) { const channelData = buffer.getChannelData(0) const PaddingLength = (Math.pow(2,Math.round(Math.log2(channelData.length)+1)) - channelData.length) let halfPaddingLength = parseInt(PaddingLength/2) const pad1 = new Array(halfPaddingLength).fill(0); const pad2 = new Array(PaddingLength - halfPaddingLength).fill(0); let finalBbuffer = [...pad1,...channelData,...pad2] console.log(finalBbuffer.length) let mfccData = Meyda.extract('mfcc', finalBbuffer) console.log("mfccData : ",mfccData); });
Result :结果 ：
[ 249.11783051490784, -90.61751411189829, -12.253094024524968, 19.88245460444982, -11.661965456271869, -14.795375019626466, 7.19298966815922, 8.362884489124907, 3.9941283332736557, -3.2158388656478287, -2.0395393071161063, -0.48849176751482837, -3.6199623273626695 ]
Why this difference .为什么会有这种差异。 How to make them same ?如何使它们相同？
1 个解决方案
解决方案1
0 2021-06-23 08:59:00
They're different because you're padding the buffer in your code, whereas the command line tool is splitting your file into multiple buffers with the default buffer size and returning MFCCs for each segment.它们是不同的，因为您在代码中填充缓冲区，而命令行工具将文件拆分为具有默认缓冲区大小的多个缓冲区，并为每个段返回 MFCC。
To make them the same, you could replicate the buffer chunking that the CLI performs in your code, rather than zero-padding the whole buffer, or you could make a CLI that zero pads the input.为了使它们相同，您可以复制 CLI 在代码中执行的缓冲区分块，而不是对整个缓冲区进行零填充，或者您可以制作一个对输入进行零填充的 CLI。 If you do the latter, we would certainly be interested in adding that functionality to the bundled CLI, so feel free to open an issue on our issue tracker to discuss if you go that route and wish to contribute.如果您选择后者，我们当然有兴趣将该功能添加到捆绑的 CLI 中，因此请随时在我们的问题跟踪器上打开一个问题，以讨论您是否走这条路并希望做出贡献。
Thanks for using Meyda!感谢您使用 Meyda！
声明:本站的技术帖子网页，遵循CC BY-SA 4.0协议，如果您需要转载，请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.