Search for a particular spoken word in audio files [closed]

Question

I have around 3000+ audio files of the same author. I need to transcribe those lectures, where the author has said about a particular word.

So I need a software solution, which will find automatically all the files where the particular word is said. There can be some variation how the word has been spoken , because the audio files were spoken over 15 years.

Free/Open Source solution is appreciated.

I tried searching , and got to know about Sphinx. But I am unable to set it up to use for my project. Any help is greatly appreciated. Please

Answer 1

You can use CMUSphinx, an open source speech recognition engine which has support for keyword spotting.

1) convert the audio to required format - 16khz 16bit mono files:

 ffmpeg -i file.mp3 -ar 16000 -ac 1 file.wav

2) build and install latest pocketsphinx and sphinxbase from http://github.com/cmusphinx

3) Download en-us generic acoustic model

4) run the detection:

  pocketsphinx_continuous -infile file.wav -hmm en-us -kws_threshold 1e-40 -keyphrase "what you need to detect" -time yes

It will print you detected keywords and their times. For the best detection accuracy you can adjust kws_threshold.