Java program to match songs (mp3) with slides (ppt) containing its lyrics

I have thousands of non-English songs in MP3s & thousands of PPTs, each displaying the lyrics of a song. I want to write a java program that would match up the correct MP3 with its PPT.

I could think of 2 strategies. I have no idea if they're even close to being implementable:- 1. Extract lyrics from song via some sort of non-English voice recognition & compare it to lyrics extracted as text from PPTs. 2. Play the song to generate a wavef开发者_StackOverfloworm. Make a "Windows Narrator" kinda program read the lyrics from PPTs to get its waveform & then try to match the two.

Notes:- 1. None of the MP3s have embedded lyrics. 2. The lyrics in the PPTs were typed in, they aren't images, so no need to think OCR.

I want to know about any other strategies you guys could come up with & most importantly, please write about any java packages that could be of any help.

Is Fourier Transform involved in any way?

Another approach: write something to extract the text from PPT and put the text plus name of the PPT file it came from into a database, text file or really, anything searchable.

Write another little app to hand a user one song file at a time; give them a way to play bits of the song, a text box to type in lyrics, and a search button that'll search through your PPT lyrics for a match. When they find a match, another button records the name of the MP3 against the matching lyrics file in your database or other file.

Hire a couple of intelligent college/high school kids to do the listening/searching.

I'm betting the project would be finished in far less time and cost than what you're considering doing.

1. If you find some good open source software for this, please post it. I doubt such a thing exists.

2. Keep in mind that reading and singing would have quite different wave forms - not to speak of the music that would have to be filtered out and the differences between voices.

Additionally, keep in mind that you might have to perform some similarity calculations since not always is the sung text equal to the written lyrics.

I'd say programming such a task is quite extensive and it would require a whole team to implement. Sure you can handle that?