Ableton Forum

This was the main reason I upgraded. I often write stuff by singing ideas just to get them down and then replaying them later. An audio to midi feature that worked properly would be very useful to me but this one doesn't seem useable at all. I don't expect it to be 100% accurate, but it's leaving out lots of notes, adding notes that aren't there, and generally making so many errors that it would be faster to just play the part.

In melodyne I can tune a 3 minute vocal track and there might be 2 or 3 spots where it places the note a half step off and occasionally a place where it glitches and drops a note down an octave below where it should be. But in general the errors are few and far between and easy to fix. It doesn't miss any notes entirely or add notes that aren't there. But in live 9 I can't seem to get it to do a simple 4 bar part without having to draw in a bunch of missing notes and basically just redraw the whole things from scratch. Am I doing something wrong?

Only tried it out for drum material but was very impressed with the results. Used a busy uptempo d&b track with a lot of sustain on cymbals but this didn't seem to throw the algo off much. A few missing hits but the groove of individual drum hits seemed pretty tight.

melocoton wrote:This was the main reason I upgraded. I often write stuff by singing ideas just to get them down and then replaying them later. An audio to midi feature that worked properly would be very useful to me but this one doesn't seem useable at all. I don't expect it to be 100% accurate, but it's leaving out lots of notes, adding notes that aren't there, and generally making so many errors that it would be faster to just play the part.

In melodyne I can tune a 3 minute vocal track and there might be 2 or 3 spots where it places the note a half step off and occasionally a place where it glitches and drops a note down an octave below where it should be. But in general the errors are few and far between and easy to fix. It doesn't miss any notes entirely or add notes that aren't there. But in live 9 I can't seem to get it to do a simple 4 bar part without having to draw in a bunch of missing notes and basically just redraw the whole things from scratch. Am I doing something wrong?

This pretty much jibes with my experience. It's all over the place, not usable for anything but the simplest melodies, which of course are easier to just play on a keyboard. I find it works well enough with slow/ very distinctly played notes, but like recording a guitar line, it is terrible at interpreting hammer on/pull off stuff or any kind of slide. It's a nice idea though, I hope they keep improving it.

just like irrelevance, i only tried it out on drum loops. yeah, its pretty neat so far. i noticed, you have to do some editing sometimes to make it sound good, because the velocities/note lengths end up being all over the place.
sometimes that can be a good thing, but sometimes its not. but for the most part, its a pretty cool feature.

Yeah, I haven't tried it on a drum beat yet. I did try beatboxing and that was all over the place as well. It seems like maybe they intended it more for ripping off stuff from other recordings and less for inputting notes quickly with your mouth. which is a shame because it's obviously possible to have a more intelligent detection algorithm like melodyne.

I wonder if it would help to do some pre-processing. I'm going to try adding distortion to my voice first and see if that helps turn it more into a machine recognizable square wave tone. kind of like a pre-emphasis on a vocoder. edit: tried it and it didn't help at all. I got quite different results but equally unusable.

Overall I've been pretty impressed with it, it catches things I didn't think it would, even while missing some obvious stuff now and then too. I think I can get better results with Melodyne, but it takes longer and always feels a step removed from the creative process, so the way it is in Live is good enough for me.

It's the missing notes that really make me scratch my head. It's just flat out ignoring certain parts and leaving silence in places where there is obviously a note. And I'm talking something recorded at a healthy level and with an obvious tone, not like it's just filtering out ghost notes or unvoiced sounds.

are y'all using audio that is in the original tempo b4 bouncing to midi?

audio to midi is also my main reason for upgrading (tho i've found other reasons to be glad i did)

sowhoso wrote:are y'all using audio that is in the original tempo b4 bouncing to midi?

yes

I use Audio to Midi, and have been since the beta started, so I have had some time to get used to it. Long story short, it works for me.

My girlfriend and I are a band, and when she comes up with a lead guitar part or keyboard riff or whatever, she sings it to me. In the old days, I'd sit there and work it out on a guitar note by note. Now she just sings into Live (usually with an sm57, for what it's worth. no need to heat up a tube mic just for this) and I convert it to midi.

We then make adjustments on the piano roll to perfect the passage, then I read off the piano roll and play the part on guitar. Or, if we're making a scratch arrangement and it's a keyboard part, I just delete the rack and drop on a nice sampled keyboard, and boom. Instant keyboard part, more or less.

It's never totally perfect, but I'd also never use something like this on stage. It has effectively speed up my workflow about 2x now that I don't have to play the guess-and-forget game where she sings and I wander around the fretboard until we (hopefully) eventually achieve the arrangement an hour later. Sure, I should be a better guitar player after 9 years, but I'm not, so there you have it.

TL;DR - I use it. Girl sings guitar solo/riff into mic, audio-to-midi happens, I read the part off the piano roll and play it on a real guitar.

any tips on how you're getting good results? does she have to make a conscious effort not to sing too legato? or does she sing a particular sound that works? I tried doing very clear Da Da Da type of lines with lots of space between the notes and it still was hit or miss.

maybe live just hates my voice

i've been impressed with my results so far. i don't have Melodyne so i cannot compare, but i've found that it works well for humming out ideas and beatboxing and also for taking a bass or guitar part and turning it into something else entirely after it's converted... it would take me much more time to figure out on my bass or... probably NEVER on a keyboard so this feature catapults my ideas and... actually lends itself to making the music i hear in my head... instead of sitting down with an idea and getting so lost in the attempts to extract whats in my mind's eye that i either

A. forget the original sound/idea that was in my head
or
B. end up with something utterly different

i can deal with B... because 19 times out of 20 thats what has happened in the past and i end up creating a tune anyway, but it sucks when A happens because those little internal songs are my inspiration for everything from writing lyrics to buying DAW software in the first place.

tldr; Audio to MIDI gives me what i want... a basic manifestation of the song in my head.

I've been practicing with it and I'm getting better results now. One thing I realized is that you need to be in concert pitch. A lot of times if I get an idea I just fire up a new live project and sing into a new track with no pitch reference. But it makes sense that doing that would confuse the algorithm. It also seems to help to really eat the mic. And obviously the simpler the better. It doesn't seem to detect fast stuff that well.

Don't forget it bases things off your transient markers in the original too, you can edit those ahead of time to get rid of extra notes and to make sure all notes are converted.

Damn Funken... Now I know how a deer caught in a headlight feels...
Beige-ish pink background. Black and grey text. Bam, pink & purple overload.
This little fella will get its best use yet ->

Ah sweet. Page 2, where retinas come to heal