Don't get your hopes up for audio (*.wav) to MIDI conversion software.

I've seen many requests in many audio/MIDI forums for such software (just like "remove vocals" software) and several suggestions but as yet have not seen anyone happy with the results of any even after getting fooled by the referral and the demos and spending some fairly large change.

The general consensus is the audio has to be one track (instrument) extremely clean and single note melodies (no chords) performed very slowly to get reasonably accurate results. Even then don't expect to edit it in any MIDI editor simply (Staff or Piano view) as it tends to represent the melody in MIDI as Pitch bend controllers.

In the other direction (MIDI to wave) you can do some very impressive things though.

MJ

MJ,
that sounds like what I thought would be the case. Seemed like an impossible task to analyse the sound of an orchestra in full tilt.

MIDI data is digital, meaning that the information sent is in the form of multiple on/off signals. It is not in the form of an analog audio signal. MIDI data can only travel on direction through a single MIDI cable. Most MIDI devices are equipped with both MIDI input and MIDI output. This means that electronic instruments with MIDI capabilities are able to transmit and also sometimes play MIDI data. MIDI specifies 16 separate channels, allowing for the control of up to 16 different instruments at simultaneously. Communication between MIDI devices is done through the passing of messages, which composed of 3-byte (24 bit) strings. These messages are transmitted at a rate of 31.25 kbaud, or 31,250 bits of information per second. When the MIDI data is received by a computer, sequencer, controller, or other MIDI device, the data is decoded and interpreted. MIDI capable instruments respond to messages according their current mode.
Midi can: Not:
Read a wav file, extract notes from it, and manipulate those notes.

Yes, a thick composition is very difficult to extract notation from, but given that I do this regularly, it works, can be notated, and is very powerful. Regardless of what you might have read in forums, mjhig, I do this regularly. It's part of archiving old recordings from Alice Fletcher, Frances Densmore, Artie Cuyman, and other artists from the early 20th century and late 19th century. Old recordings have some issues and artifacts, but anything newer than about 1950 is very clean and easy to extrapolate notes from. Again, a full orchestra isn't possible, but nailing a banjo out of a track or nailing a violin out of a track of a quartet for instance, is very doable with few errors. It's not remotely the same as a "remove vocal" which is harder for a number of reasons. It's tracking a specific set of defined frequencies. Removing them leaves a huge hole, and vocals have attributes that few instruments can voice. Some instruments, particularly polyphonic instruments are tough because of the numbers of frequencies, but it still is quite doable. Don't believe everything you read, particularly in forums.

Spot,
My understanding (based on searches from a couple years ago; I know, tech changes fast) was also that this sort of extraction was monophonic only. This is intriguing; what software do you use to do this, and is that a major coup of that particular software, or does its competitors do this as well?

The problem with forums is you get the deicated crowd who just lik being there because enjoy helping others and the ones who have a problem.

You never hear from the people who just opened the box, installed the software and went to work without any problems at all So forums can give you a very distorted view of a product unless you carefully evaluate each post

pfs is great, if you can still find it, Digital Ear is the best for extracting the emotion of a performance. In fact, since I can't write my flute scores well for others, I use Digital Ear to extract my flute's expressions. It is a monophonic extractor.
Akoff is a good extractor too. I reviewed Akoff as shareware several years back for REP magazine, I think it's still around too. I'll admit they aren't perfect, nothing in this realm is. But with a good recording, a little EQ on the wav file before extracting, it's quite easy to get very close 75-80% of the time.

I fail to see much difference other than a positive spin, between what I said and you said Spot;

>>>>The general consensus is the audio has to be one track (instrument) extremely clean and single note melodies (no chords) performed very slowly to get reasonably accurate results. Even then don't expect to edit it in any MIDI editor simply (Staff or Piano view) as it tends to represent the melody in MIDI as Pitch bend controllers.<<<<

I certainly don't believe all I read but in this case the issue is what's acceptable results for one, is not for another.

My goal was to be sure the readers who are interested in audio to MIDI conversion approach it with healthy skepticism. I'm sure the majority of potential users aren't expecting to do much more than point the software on "any" piece of music, press "OK" and expect that all the instruments are broken down to separate MIDI tracks with all it's notation, including chords, drum voices, the whole nine yards. They should be aware, nothing like this is the case.

MJ

Lots of useful stuff here, but to get back to the core of my question all I really want to do is:

Inport a wave file that's some song. CHANGE the sound of it meaning, if it was originally just a piano melody only, I ADD some strings, horns, whatever sounds good via MIDI synthezied effects either by a sound card and/or the software. No keyboad entry, no writing of music, no extraction of vocals, none of that. Then blend it all together ending up with a new single WAV file. Exactly what I did about ten years ago. Jut can't remember how I did it or what software I used.

Dude, are you sure you started out with a wav file? I just don't see how you did this so easily back then, besides doing what all has been discussed so far. Are you positive you weren't importing a midi file, then simply changing the patch it uses? Even now, without some sort of extraction as has been discussed, I don't see how you change the "sound" of a wave file besides using FX and whatnot.

Obviously, this is a piece of cake with an actual midi file of a song. But if you're certain you were importing wav files and not midi files, I have no idea how you did this.

Yep, I wish I could remember the name of the little application that did it and exactly what I did. I sometimes can't remember what I did yesterday let alone 8-10 years ago. Horror of horrors,maybe it wasn't a Windows application maybe in ran in DOS but I'm sure it had tracks where you could import a wav on one track and do MIDI on others then combine the whole thing. Maybe it didn't do things some would require but for what I wanted it worked really nice.

OK, that you can do real down and dirty with a free app called the Jazz ++ MIDI sequencer for free (Google for it), maybe with one of the demo Acids and very well, very cheaply with Cakewalk's Home Studio 2004 (under $100 I believe).

MJ

Acid can do this VERY well. Drop the wave file in chopper, pick up the beat, then add ACID loops or MIDI information. Render out as a wav file

When I read Billyboy's first post I thought he wanted to sweeten some audio tracks by adding a couple of MIDI tracks, like strings or percussion. This (and adding audio tracks to MIDI compositions) is fairly common. I see that's not what he meant. What he remembers doing in the past I'm not sure. Perhaps they were MIDI tracks to begin with?

In any event, the turn this thread has taken is interesting. Audio-to-MIDI transcription (extraction is a misnomer) has been a musician's pipe dream for a long time. The conversion utiilties I have tried do a respectable job with an isolated diatonic solo instrument and can get within a half-step much of the time. Maybe someday these programs will be able to recognize timbres and pull the fundamentals from the sea of harmonics. I do all of my transcription with headphones, Sound Forge, keyboard, #2 pencils, and staff paper. Call me old-fashioned.

As for the "state of the art" of polyphonic transcription, read this article and others, and draw your own conclusions.
http://emusician.com/ar/emusic_ims_intelliscore_polyphonic/index.htm

A freind of mine tells me he has an freeware app that even lets you just whistle or hum into his crappy PC mic and it'll put the notes on a staff and create a midi track. He's used it to create his own ringtone for his mobile.

I'll se if I can get him to find the name of it, he's one of those people who always 'used to have' things.