Meyda 特征提取的数据在命令行和通过 nodejs 不同

Question

我尝试使用 Meyda 提取音频文件的特征。 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

1. 命令行

   梅达 1600153162.63571.wav mfcc

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 9,5.45750656573637,5.059966218569768 23.469876801827922,20.984500104582974,15.80549272827184,10.064116650359527,3.957435073588796，-1.75506262168123，-5.904087102697238，-8.29320895086182，-9.037630914839253，-8.230609293960022，-6.604159793447832，-4.8505887091085365，-3.2390627275186175 29.14206006610766,25.50517483703385,19.279159148659456,13.806566615476791,7.715092034538,2.2364700767576764， -0.8129542016331001，-2.16772108652068，-2.6257181917727093，-2.1347711722007845，-1.0689231846996041，-0.6059541306434987，-0.9955650805848651 12.251538716256618,9.479093353756676,5.26726325506769,2.7893618666871762,0.10288164291841367，-3.121859159330943，-4.446188894370852，-3.5959072906303837，-2.5699522942015007，-1.8300687753240275，-0.7140618177046422,0.3644613187160053， 0.8353185065087604 18.42772721964866,15.990459244575096,11.993313228734504,9.088259023158539,6.172351883886699,2.6632784721745213，-0.2683186290158647，-1.948275236251193，-2.793271841187546，-3.305172382572076，-3.6707865100252994，-3.638999985131 219，-2.863263822827715 15.915610973257571,13.737073178518493,9.801934620786648,5.779403187264773,1.358952752885622，-2.5086411759816647，-5.187471901991046，-7.0034704192058985，-7.837140523422115，-7.437282553994893，-6.058991955183858，-4.360246975428363，-2.680629805396066 15.75090683856979,13.497089898444463,9.489154494388005,5.578283842168272,1.4567586737610156，-2.141429668615224， -4.380728646793593，-5.364310654901947，-5.049848052960211，-4.214979808517224，-3.9745754014376224，-4.076571890925421，-4.020714303609834 23.361729244701564,20.83103985201436,16.592090051349242,12.885538444506535,8.373777248668826,2.9768227085972145，-1.399644048138661，-4.090450595665753，-5.574775136428788，-5.787457864519816，-4.693018758118676，-3.100993535744204， -1.5625546377399475 12.793707716744393,10.197116076581045,5.819041457288568,2.4679725003955064，-0.33041998215142393，-2.607817847380143，-4.086120143731154，-4.779218992884254，-4.374465394096245，-3.118504640745296，-1.9666020102081847，-1.5140789377241373，-1.513064 7010950313 3.2732387410942465,2.2453286293698977,0.9749483649128511,0.472715919767034,0.11249067642678866，-0.19927487640318992，-0.24286230087509772，-0.3645480814147444，-0.5060378572115204，-0.4793996411313405，-0.45934442588699886，-0.4680761055711004，-0.2967506177492105

2. 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); });

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]

为什么会有这种差异。 如何使它们相同？

Answer 1

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。

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。 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。

感谢您使用 Meyda！