I'm doing research on how we can do matching between 2 audio files using Android to see if they have the same content or not. After doing lots of research, I have concluded that I must do 2 main things:
- FFT
- Correlation
Am I on the right track?
1
$\begingroup$
$\endgroup$
-
$\begingroup$ How are the two files similar? Are they identical enough where one can do a bitwise compare? $\endgroup$ – hotpaw2 Jul 6 '13 at 22:10
-
$\begingroup$ in similar i mean they have the same words and phrases ... you know? like to see if 2 people said the same thing. @hotpaw2 $\endgroup$ – nada Jul 6 '13 at 22:12
-
$\begingroup$ If you application is speech, you should look, specifically, into the extensive speech recognition literature. This is not my specialty, but my understanding is that FFT and correlation are generally considered too CPU-intensive to be useful. en.wikipedia.org/wiki/Speech_recognition#Algorithms $\endgroup$ – Bjorn Roche Jul 7 '13 at 0:23
-
$\begingroup$ I tried doing speech recognition using google service ,,, but i was not accurate! (in my language) @Bjorn Roche sooo im trying to use the DSP approach. so you think FFT wont work? $\endgroup$ – nada Jul 7 '13 at 15:49
-
1$\begingroup$ Correlation will identify two waveforms with similar shape. It won't help you find similar words. It is usually implemented using fft for speed but doesn't have to be $\endgroup$ – endolith Jul 7 '13 at 17:55
3
$\begingroup$
$\endgroup$
It really depends on which application you have...the algorithms for speech recognition are different from speech indentification for example. Based on what you need, just look for an open source library for Java. Getting into the theory is definitely desired but as a first step you can just look for a software block that has inputs/outputs, later you can drill into it.
Here is some info on speech recog: https://stackoverflow.com/questions/3018418/open-source-simple-speech-recognition-in-c-in-windows
-
$\begingroup$ my application is going to be an app which does matching between 2 audio files. And i already started! and i have an app which records 2 voices and im working on the FFT right now. i just need someone to assure me that what im doing is correct,,, because some people tell me im lost! :O $\endgroup$ – nada Jul 7 '13 at 15:18
-
1$\begingroup$ but what are you trying to do? can you describe the application? $\endgroup$ – TommyG Jul 7 '13 at 23:04
0
$\begingroup$
$\endgroup$
cross spectral match 2 audio wav
