KMan, I am glad to report that your problem is solved...


I am confident that you are able to clone your own snare drum playing, can’t you? Playing along with your recording? Sort of à quatri mani like the Bridges brothers? (The Fabulous Baker Boys)

If that’s the case, here’s your solution:

Boring but Brilliant Theory: [Best Paper Award]

Dropped-my-Jaw, Blew-my-Socks Practice:


ps: I have been offering my services to the Audacity folks, to work on this stuff, but they have so far dismissed me. See cartoon below.

Can you explain what you are talking about please.

Yes we know there is software that attempts to do source separation, and I’ve seen software similar to that in your MP4 where you can draw a shape around the voice in the spectrogram and isolate or separate it with some success.

Are you the developer of such apps? What are you offering? Where did you offer that?


Sure. It will be my pleasure.

Sometimes, in some initiatives, the environment is not too conducive to experimenting, looking for new stuff in a rapidly changing state-of-the-art.

One example is when a newcomer gets a response like the one below.


None of the other Forum staff disagreed with that “personal opinion” of mine (they would not be afraid to disagree if they did so).

This is the Audacity Forum, not the iZotope Forum. I suggested where you could create an iZotope Forum.

Again, what are you offering, please? Are you a developer looking to implement Proposal Source Separation for us? Then I for one would welcome you with open arms :smiley: but you should be talking about your ideas here: and not on Audacity Forum. Developers don’t hang out on the Forum.

If you are offering to help our users with whatever commercial plugin you are doing source separation in, how does that help our users who don’t have several hundred dollars to spend? Is it an iZotope plugin you’re using? We’ll permit you to say what it is :wink:

If you want to receive users’ complete tracks, process them in the software you use then send the tracks back to the user, then as you point out, this Forum is not set up to be that file transfer service. Again my personal opinion, but I don’t think the Forum should be such a service because file storage apart, it’s our bandwidth you are using up at our expense, and this is free software.


Also what does have to do with this? Your PDF is not about Sphinx as far as I can see. Do you work at CMU?

If you are not a developer, what features are you requesting in Audacity? You can ask for features here But please be specific.


(1) THE state of the art, most productive approach in sound decomposition is Non-negative Hidden Markov Modeling:
(The PDF must be downloaded first, in order to listen to the sound clips inside it)

It is the theoretical base for Sphinx (*) and for the demo of the next Adobe Audition version that I posted.

(2) Sound Decomposition is the philosopher’s stone of Sound Processing.'s+stone

(3) Sphinx is Open Source and so is Audacity.

(4) Audacity could take advantage of decades of research work and code (millions of dollars worth), courtesy of Uncle Sam.

In any event, in my initial attempt, the Sphinx spokesperson (the one and only, sadly) confirmed my suspicions (**), mentioned some buzzwords and threw the door on my nose. :frowning:
(Note: Travis Banger is my alter ego)


(*) Sphinx has forced other Speech Processing initiatives to simply shut down. Reminds me of when TCP/IP killed all competitors.
(**) The only way to do proper voice NR (which will be renamed Noise Elimination soon) is via voice recognition techniques (HMM)

I hired a guy (top of the line, pointed me in the direction of iZotope RX4, before I discovered Audacity) who used that mentioned Photoshop technique to sort-of denoise the audio below.


(Noise begins at minute 1, until the end)


It took him almost a month.

That is when I concluded: “What needs to be done here is use/develop a system that recognizes voice as such and dismisses the rest”.

Next, I read a post of yours saying “We are not interested in doing that” (or words to that effect).


So, Sphinx cold-shouldered you and I now see that you posted in and got no response. That would be because (a) no developers read this Forum regularly (b) you framed your question as a joke, and I guess people therefore aren’t clear what you are trying to achieve.

Could I suggest you may do better if you lay out your aims clearly to begin with - what you are trying to do and who you have approached?

I moved that thread to feature requests.

So what are those feature requests? I thought it was that you wanted sound separation for voice removal and isolation.

Now it seems to be whether we could use the CMU Sphinx techniques in Noise Removal?

Or is it both - Sphinx could be used to remove/isolate voices and denoise audio?

Is Markov Modeling useful for processing music, or only speech?

We have improved Audacity Noise Removal to some degree in 2.1.0 but I for one would accept that it is still not as good as in a quite a number of other applications. So ideas and new departures are welcome, but this Forum is the wrong place to post.


You proposed, explicitly, that we should host a forum for iZotope users here. I gave my opinion that it is not appropriate for us to do that.

I also said that personally, I have no interest or motivation in using iZotope tools (or by implication, helping others to do so). That will be the same for most of us who work on Audacity, I guess. We are experienced in and offer support for Audacity. We are not experienced in using closed source expensive software and do not offer support for how to use other software (free or commercial) because we do not have the time to do that.

Your post did not say “Can Audacity use/develop a system that recognizes voice as such and dismisses the rest?” If you had asked that question instead of expecting us to subliminally know your reasons for asking that question, you would have got a (somewhat) different answer.

If you had explained exactly what this is all about then you would have got a much different answer.


I did? When?

See post in question below.


My apologies for my lack of accuracy. The one who actually said “We are not interested in doing that (or words to that effect).” was not you, my esteemed Gale.

It was the always helpful Koz.

See below.


I must have missed that post. What are “your services” and what contribution to the Audacity project have you offered?

Then why pose a question asking if there were iZotope users here? Wouldn’t that be like creating a forum within a forum?

If you’d said why you wanted to contact iZotope users (because you want to implement “Non-negative Hidden Markov Modeling”) then I would have been positive about learning more. I have no interest in supporting iZotope users on this Forum, which is what it looked to me like you were asking.