Hi there. I am in need of a bit of help. I’m working on a project where I have to extract specific sounds out of a single audio track. I have two tracks
Track A: The original. Several short pieces of music, knit together. This track contains several sound effects unrelated to the music that I have to extract.
Track B: A reconstruction of the music of track A, nothing more.
I intend to use B as a ‘negative’ to cancel out the music in track A. However, I have encountered two problems
The volume level on Track B is louder than track B. About 60%
Track B has been ‘reconstructed’, so to speak. As a result are some gaps in timing between the three or so little pieces of music. You don’t really hear it while playing, but by the end of it track B is almost 3 seconds longer than track A.
I was planning to cut Track A and B into little pieces so the timing of the music would be easier to match up. Still, it’s quite hard. Are there ways to do match these pieces up automatically? About the volume, I measured the highest db output of both tracks and tried equalizing that, but it just wasn’t anywhere near accurate enough.
Accuracy is the key to all of this. Without it I cannot use track B to cancel out the music in track A.
Does anyone here have any ideas on how to best approach? I’m sorry I can’t post the tracks here, it’s all copyrighted material.
Actually, it’s nearly impossible. Nobody can resist messing with the instrumental minus the voices and that kills you when you try to use one to cancel the other.
There were danger signs at problem one. Right away, one is not the other with pieces missing. By the time you got to problem two, that’s the end of the show. Any production you do to one without the other will prevent you from achieving cancellation. As you’re finding out, cancellation only happens under one, very specific, critical set of conditions. Change any condition however little and you can pack up and go home.
You didn’t say the phrase that really causes problems. Are both tracks MP3? MP3 causes sound damage – always – and the damage changes with content. So the vocal track has a different set of compression artifacts than the instrumental does.
I don’t know of an instance where anybody has ever made this work.
Koz, thank you for replying. Neither track is mp3, but track 8 has been through a high-quality video codec in the past.
I don’t know of an instance where anybody has ever made this work.
Koz
That really seals it then. A pity. I was hoping to get them at least some kind of ok-ish quality. And that perhaps enough people had done that for there to be more tools to try out.
It’s not the end of this project, I actually started out by recreating all of these effects from scratch. At some point I simply started wondering if there wasn’t an easier way to do this. I mean, you hear about interesting advancements in musical software all the time these days. Music recognition software for Android phones, with which you can record live anything you want; it will tell you what tune it was and who played it. Melodyne, a plugin for music sequencers that makes raw audio formats of any kind completely sequenceable, note by note, even at LP record quality… the sky seems to be the limit.
It was worth a shot. Thanks for letting me know the score.
