Audio to Score conversion/transcription

I’ve tried - I really have… but not had much success with either Nurotron’s AudioScore or AnthemScore.

I’m prepared to do some work configuring and experimenting, of course.

But perhaps it’s asking too much to have software - even on a fast machine with commercially-produced CDs (converted to MP3s etc) - accurately produce MIDI or MusicXML (transcriptions) from anything other than a single instrument?

Anyone had success with anything ike this, please?


It is, I think, basically asking too much. You would probably find that using software designed to assist you in the transcription process, rather than software that attempts the transcription itself, will be more successful, but you will have to do more work yourself, of course. Transcribe! by Seventh String Software is, I think, the preferred solution for this.

Thanks, Daniel. Thought so :slight_smile:

Yes, I have Transcribe! I’ll take another look…

Converting audio to midi (which can then be converted to score) is a very difficult programming problem which people have been working on for decades - and with very limited success. The crude basics are simple enough - you mathematically analyze the audio and figure out which frequencies are in the source. But it’s the details which are immensely complicated - how do you determine which is the fundamental note vs. the overtones/harmonics - how to handle even the slightest shifts in tempo, etc, etc, etc,

I’ve recently been mucking around with two different “audio to midi” applications - Melodyne & Widi - with rather limited success. Both advertise the ability to handle polyphonic material. I my experiments, Melodyne did a somewhat crude transcription of a piano piece - but it needed a lot of work before it could be made accurate. Widi’s conversion was completely nonsensical - but maybe I was doing something wrong.

That said, you can download a 30 day trial version of Melodyne, and you can buy a 30 day trial version of Widi for $30 US.

Thanks very much, eheilner!

I suspect the whole thing is unrealistic. Some software does claim to do it, though. Like OCR.

As someone who devoted 4 years of a PhD on this very subject, I am extremely sceptical of many of the claims made by commercial products. Audio to score for polyphonic music is Hard. Note that even if you have audio to midi then that’s still only half of the problem, and tidying up the midi for import to the score is also very much non-trivial. I would advise anyone looking at such a product to try a demo version where possible, and try on the sort of content you usually deal with. That will help you to manage your expectations about how useful the tool will be to you (if at all) and whether the subsequent tidying up takes longer than if you had entered it by hand.

I have spent most of my career doing “take downs” (nearly 1,500 and counting) for London Orchs and my own MIDI file company. I have tested nearly all the software on the market and they all fail quite miserably if tasked to anything more than a simple monophonic line.

This means that your own ears will be the best tool you have. Slowing down the music slightly does help, but don’t do it too much and a good EQ plugin and phase cancelling module can also be useful.

If you would like any further help please drop me a PM.

Good luck.

Thanks, Paul and TopDots!

I hear you loud and clear - unlike the software does/can.

I’m going to try slowing down with Transcribe and/or then entering manually. It will improve my ears too.

Everyone’s advice much appreciated.

I have to emphatically second that. Even though I detest doing take-downs (although there is always a certain sense of victory in hitting full orchestra ones out of the park), they are part of the business and no software can help. It also gets quite a bit easier with time. I’ve tried audio transcriptions tools too, but I’ve always been extremely underwhelmed. The two cabbage leaves on either side of one’s head are the answer …

Thanks, Claude; I’m convinced!

There’s an article here (disclaimer - it’s written by me) about how even if you have software that can perform note detection reasonably well, that’s just the beginning of your transcription problems.

Thanks, Andy. The style and tone (no pun intended) of your documentation always makes it quite clear. I’m dizzy :slight_smile: