测量何其相似的两个短的音频算法简单 [英] Simplest algorithm of measuring how similar of two short audio

问题描述

问题是寻找任何开源或简单的实现来衡量如何在iOS适用的两个音频相似。

The question is to look for any open source or simple implementation to measure how similar between two audios on the iOS application.

简单地说，音频可以重新由1-D矢量psented $ P $，以计算一维矢量之间的距离。但音频的长度会有所不同，因此需要一些pre-处理等。

Simply speaking, audio can be represented by 1-D vector, to calculate the distance between the 1D vector. But the audio length will be different, therefore need some pre-processing etc.

期待在这里得到一些线索，谢谢

Looking forward to get some clues here, thanks

推荐答案

可变长度的两个序列之间的相似性可以用DTW被有效地计算：

The similarity between two sequences of variable length can be efficiently calculated with DTW:

http://en.wikipedia.org/wiki/Dynamic_time_warping

此算法简单实现自己，还有维基页面上链接的相当多的现有实现的。

This algorithm is simple to implement yourself and there are quite many existing implementations linked on the wiki page.

简单地说，音频可以重新通过1-D矢量psented $ P $，

Simply speaking, audio can represented by 1-D vector,

这是合理的分割帧上的音频，使之成为在那里的每一帧你的价值（功能）阵列功能的2-D矢量对应不同的频段。如果你要处理的音乐，对每个帧的FFT是一个好主意，语音，最好是计算<一个href=\"http://practicalcryptography.com/miscellaneous/machine-learning/guide-mel-frequency-cepstral-coefficients-mfccs/\"相对=nofollow> Mel频率倒

It's reasonable to split the audio on frames and turn it into 2-D vector of features where for each frame you have an array of values(features) corresponding to the different frequency bands. If you want to deal with music, an FFT for every frame is a good idea, for speech, it's better to calculate mel-frequency cepstrum

同样，你可以使用梅尔频率的功能很多现有的库，其中一个是语音识别工具包 CMUSphinx

Again, you can use many existing libraries for mel frequency features, one of them is a speech recognition toolkit CMUSphinx

这篇关于测量何其相似的两个短的音频算法简单的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！