Thank you very much for this!!!

[sihirar]

Hi all

I just wanted to thank the Audacity developers for developing this wonderful piece of software. I've been writing a program for the last couple of months that automatically mixes any songs I feed into it, and Audacity plays the processing role in it.

The goal of my program is to automatically generate a mixtape in a few minutes with any songs that you chose, and make it sound like it was mixed by a human (so not like the Windows Media Player autofade-like gimmick). I got the idea at New Year's Eve. I was behind the turntables mixing songs at a family party. Everyone was having fun, and so was I because I like DJ'ing, but I kinda felt I missed the whole being together with your family experience. Everyone was mingling with one another and having a good time and dancing on the dance floor. Meanwhile, I stood there mixing music and looking at everyone having fun. And that's when I thought; hey, there should be a program that replaces me so I can go and have a good time too! And so my program was born a month later.

As of today, it's far from finished but it can already do a long list of things. For example, it can automatically look for the best moment to start mixing in the other song (it can detect when someone is singing so it won't mix two songs where they're both singing thereby clashing the vocals together), it automatically loops when the first song is too short to mix smoothly with the second song, when given a large list of songs it finds out the keys of all songs and mixes the songs harmonically based on the key they're in, it matches the tempo of all songs mixed, it adjusts the volume of the tracks mixed so that one track is not louder than the other and much MUCH more.

I am planning on releasing the program sometime but there is still A LOT of things to fix which I don't know how to fix yet and which I'll try to figure out in the next few weeks/months. As of now, it's working flawlessly with electronic music and the like, but rock or pop songs don't mix quite well yet and that's what I am looking to fix before releasing.

To give you an idea of what it can already achieve, I'll include some samples of some of its work.

Sample 1 - mixing a non electronic song with an electronic song, the first success it had with non electronic music!
Songs mixed: A: Ellie Goulding - How long will I love you / B: Duke Dumont - I Got U

http://vocaroo.com/i/s1fUlKyJTTFL

Notice how the program notices that song A is about to end and saves the day by looping it until track B offers a good chance to fade song A out smoothly.

===============

Sample 2 - mixing two electronic songs (the genre it currently mixes the best)
Songs mixed: A: Redfoo - Let's Get Ridiculous / B: Deorro - Bootie In Your Face

http://vocaroo.com/i/s0EvtJlStt3J

===============

Thank you VERY MUCH for this crown jewel called Audacity!

Many greetings from Belgium.

[Gale Andrews]

Thanks Sihirar for your interesting description.

Can we hope that your finished application will be open source under GPL, or under a licence compatible with GPL?

If it isn't, you'll need to make special arrangements to install the Audacity engine separately from your application, so that the Audacity open licence is distinct from your own EULA.


Gale

[sihirar]

Hmmm.. I haven't really thought about that yet, but I'll leave that for later.

Ironing out the bugs and making sure that it works optimally with any type of music is what I'll be focusing on now.

I noticed that the link to the second sample is wrong by the way, it should be http://vocaroo.com/i/s1tYisgUoUJK

[kozikowski]

it can detect when someone is singing

That alone is worth looking at the code.
Koz

[Gale Andrews]

it finds out the keys of all songs

For polyphonic material, about 50% success is the best that most current apps will achieve.


Gale

[sihirar]

it finds out the keys of all songs

For polyphonic material, about 50% success is the best that most current apps will achieve.


Gale

Well, it does not use an application for that. It finds the key by using the API of EchoNest. I don't know how accurate their database is and whether or not they use an app to find out or get the information from the artists themselves, but judging from what they have achieved with audio I have great trust in them.

it can detect when someone is singing

That alone is worth looking at the code.
Koz

It's actually a lot less complicated than you may think it would be. It does not really 'know' when a voice starts or stops singing. It just uses timed lyric sheets. For example, when the duration of a song is 4:00 and the last phrase of the song is at 3:15, it will wait until at least 3:25 (to make sure the phrase has ended) before mixing.

The downside to this is that in order for this feature to work optimally, songs used should be purchased songs, official songs. Songs from YouTube or any other source might be off time-wise thus rendering the feature inaccurate. Also, a timed lyric sheet should already be available for the song that you are trying to mix. If no one has yet bothered making one, I have to either make one myself which takes a lot of time or just not use the anti vocal clashing feature. So this is still one of the points that I need to work on, in the future I will be looking for more reliable and more independent ways to find the vocal parts in a song.

[Gale Andrews]

It finds the key by using the API of EchoNest.

Thanks. Well EchoPrint ( http://echoprint.me/ ) is open source but the EchoNest API apparently isn't. So it doesn't look like your app could be open source.


Gale

[sihirar]

It finds the key by using the API of EchoNest.

Thanks. Well EchoPrint ( http://echoprint.me/ ) is open source but the EchoNest API apparently isn't. So it doesn't look like your app could be open source.


Gale

That's awesome, I didn't know about EchoPrint, I'm going to look into that.

[Robert J. H.]

Your application will obviously be working with a database, rather than a own deviced feature-extraction, won't it?
I would have been very surprised if you had implemented all the algorithms that your features suggest.
It is of course perfectly legitimate to go this way.

[sihirar]

Your application will obviously be working with a database, rather than a own deviced feature-extraction, won't it?
I would have been very surprised if you had implemented all the algorithms that your features suggest.
It is of course perfectly legitime to go this way.

It only uses a database for key finding and knowing where the vocals are.