帮助：Speech API：本地热词检测 [英] Help: Speech API: Hot Word Detection locally

问题描述

我开始实施NodeJS

I started a NodeJS implementation

• 使用SoX捕获音频流，
• 使用Oxford for Speech2Text
• 使用法语Soundex进行句子清理
• 然后Levenshtein找出匹配的"语法"

它看起来工作得很好但是我真的需要一个算法在我的流中本地匹配一个像"Hey Cortana"这样的热词。

It seems to work very well BUT I really really need an algorithm to match locally in my stream a hot word like "Hey Cortana".

我在寻找每一个在哪里和不知道如何做到这一点:-(如何"嘿Cortana"，"Ok Google"，"Hey SIRI"工作???快速傅里叶变换？

I'm looking every where and don't know how to do that :-( How "Hey Cortana", "Ok Google", "Hey SIRI" works ??? Fast Fourier Transform ?

微软或任何人都可以提供样品或提示或入门吗？

Can Microsoft or Anyone have a sample implémentation or tips or getting started ?

非常感谢！

推荐答案

看起来我正在寻找的是" Keyword Spotting "。但我找不到简单的实现......只有研究论文。

It seems what I'm looking for is called "Keyword Spotting". But I can't find simple implementation ... only research paper.

• CMUSphinx有一个实现h使用本地语音引擎的C代码中的idden
• 确定Google是一个深度的neaural网络，培训了40,000个样本

我不明白人们怎么做继续听牛津听？或者如何"嘿Cortana"实施了吗？

I don't understand how people do continous listening with Oxford ? Or How "Hey Cortana" is implemented ?

SoX允许我剪切语音样本但是如果我不过滤更多的话，仍有大量样本和请求到牛津...

SoX allow me to cut speech sample but there is still lots of sample and request to Oxford if I dont filter more ...

Microsoft是否提供了一种  K eyword Spotting，例如"Hey Cortana"。对我们来说？

Does Microsoft provide a kind of Keyword Spotting like "Hey Cortana" for us ?

这篇关于帮助：Speech API：本地热词检测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！