ARA 2 and Melodyne "Fuzzy Syllables"

[RexRed]

I keep thinking that I am not very pleased with the way ARA 2 converts my tracks into Melodyne.

Is ARA downsampling the files?

Is there a way to tell ARA 2 to do a better job converting the waves into Melodyne?

Sometimes you can't move the pitch or formant on a syllable (especially a transient) even a tiny bit and the wave gets all "fuzzy sounding". You are stuck with the way it is.

It seems to be the way ARA converts waves. It doesn't matter if the wave is 48 khz or 96khz and 16 bit or 24 bit. 

I am recording in 48khz 24bit, you would think that raising the khz or bits would allow you finer adjustment on the pitch and formants without the sound going all "wonky".  

 

[msmcleod]

ARA doesn't do any conversion, it's just an method to allow Melodyne (and other ARA region effects) direct access to the WAV file on disk.

The whole point of ARA is there's no conversion.

If any conversion is taking place, it will be in Melodyne itself.

 

[treesha]

Have you tried comparing the different algorithms in melodyne to see if another algorithm would handle the material better? I have found one gives better results than another at times, but sometimes the edit is just not going to sound ok. 

[mettelus]

In addition to the above is to test with a wave file in Melodyne stand-alone version. The algorithm chosen can make a big difference and you sometimes need to split blobs. Which version of Melodyne are you using?

[RexRed]

MSMcloed, Treesha and Mettelus thanks for you input into this discussion. 

This is in response to all of your helpful advice,

I am using the latest version of Melodyne Studio.

It is good to know that ARA is not resampling my waves but what is odd is if I remove Melodyne from some wave files it does not restore my exact takes...

It leaves behind what looks like what i would get if I bounced multiple clips to track. So if ARA is not doing this then Melodyne is resampling my waves.

About the algorithms in Melodyne there is only one algorithm choice for what I am using Melodyne for (melodic).

It only stands to reason that something is converting/resampling the Cakewalk takes/clips into another file. If this was not the case when you removed Melodyne it would restore the takes/clips exactly the same way they were originally.

It is this resampling process that I would like to know more about.  It seems to be only one flavor and Melodyne only works with that kind of wave file. If this was not the case then higher project resolution would yield better pitch and formant editing fidelity.

Please correct me if I am wrong.

The issue I am having is I bend my notes up I drift up to my notes during the transient phase of singing. This is the area where I need better resolution in the resampling process. Melodyne does convert these okay but there is no deep resolution there to allow for pitch and formant editing. This does not always happen but sometimes even the slightest  pitch or formant editing of the transient makes the quality of the transient sound wonky or fuzzy.   This has been happening for years all the way back to VVocal over the course of three or four operating systems and many different computers.

I would like more flexibility on what kind of resolution, filter or frequency response is being used within Melodyne to convert these waves.

It is apparently not editing them in place or removing Melodyne would restore them back to their original state. It seem quite probable that Melodyne is converting them to some sort of generic format that it can understand. This conversion process I would like to have some flexibility with.

I am sorry if I am not understanding how this works...

Edited March 27, 2021 by RexRed

[treesha]

Just a quick reply when i have melodyne on a clip i do not want to edit but want to use for reference for another track in melodyne that i do want to edit, when i remove the melodyne from the reference track that i did nothing to in melodyne, it is restored to its original state just fine, which if I understand your situation is not the case for you? 

[RexRed]

Thank you for your response Treesha, drag and select several clips at once in a track, then right click on them and select region effect > Melodyne.

Notice how Melodyne puts them all into one Melodyne instance. Then right click on that instance and select region effect > remove Melodyne.

You will notice you are not left with several clips but one solid clip. In order for Melodyne to make them into one solid clip, my reasoning is, there has to have been some sort of bounce.  

When I make a vocal, I record ten to twelve takes or more of the entire vocal in a track, then I comp each part I want consecutively in the track, then I delete muted takes and I prefer to move all of the takes manually into one take lane so there is no resampling . Then I select all of the takes in that one lane. Then I right click and add Melodyne to the whole region of clips, I leave Melodyne and do not freeze the track so I can always go back into the song project and tweak the vocal in any place I notice a problem in timing, pitch etc. I save the project prior to comping with a new name, then I save it again with a new name after I delete the muted takes, save it again with a new name before adding Melodyne to all the clips and I save it again with a new name after adding Melodyne to all of the clips.  I never freeze my Melodyne tracks. I do freeze my midi tracks almost always but rarely if ever bounce my Melodyne tracks.

This melding of clips together in Melodyne I would refer to as resampling or bouncing. The fact that Melodyne can undo all of its edits and provide you with the original wave form does not imply to me, in my mind, that the wave was not resampled initially. 

Edited March 27, 2021 by RexRed

[treesha]

I had to smile about your saving projects with different names often, i do the same. I never have selected separate clips in a track to put melodyne on all at the same time so i will try it tomorrow. I either do one clip or the whole track. I have tried to use process  - audio - gain on a few separate clips in the same track at one time and gotten very unexpected results so I don’t do that anymore. Thats been my only try to process separate clips weird thing. 

[RexRed]

Hello Treesha, again thanks for the comment, I rarely  if ever use Melodyne on a single vocal clip, I usually apply it to all of them at once on a track. The only problem you can have doing this is Melodyne does not like it if the clips overlap one another. Go through them beforehand and make sure they do not overlap, they can be in separate take lanes when you apply Melodyne globally but they should not overlap or be doubled. You can also comp Melodyne, and delete a section within a Melodyne instance. You can place another clip to fill the gap and apply Melodyne to it (make sure it does not overlap).

Sometimes Melodyne will still see the old blobs from the deleted comp section. Just save Cakewalk close it and reopen it and the old blobs will be gone.

[treesha]

 

 

15 hours ago, RexRed said:

 drag and select several clips at once in a track, then right click on them and select region effect > Melodyne.

Notice how Melodyne puts them all into one Melodyne instance. Then right click on that instance and select region effect > remove Melodyne.

You will notice you are not left with several clips but one solid clip. In order for Melodyne to make them into one solid clip, my reasoning is, there has to have been some sort of bounce.  

 

Ok I tried putting melodyne as a region fx on 4 separate audio clips both all in 1 track or spread out over different take lanes. I didn't edit anything, then did remove melodyne and yes I am left with one solid clip of the 4 pieces combined.

So then I made a project folder to experiment on, I put  a wav file from cymatics into cake project at 441 24. (First I confirmed it is 441 24 in soundforge.) In the project folder cymatics wav shows in the audio folder. Then I cut up the cymatics wav into clips and spaced them out, put melodyne on the 4 clips of the cymatics wav file and looked in the audio folder again and a new file appeared called audio (bounced, 3).wav. I put that bounced file into soundforge and it says it is 441 32 bit*. After taking off the melodyne, the bounced. 3 when put into soundforge again still says is it 441 32 bit*.  I have my cake option render bit depth set to 32 bit.

So then I started over in a fresh project and changed my cake render bit depth to 64 bit. Put in a different wav file, checked in sound forge to confirm it is 441 24, the cut it up and spaced clips apart, put on melodyne then took off melodyne, got the combined image in the track, checked the bounced file that is in the project folder's audio folder in sound forge and it now says this bounced file is 441 64.  So changing my render bit depth changed what this bounced file is. 

Somebody with more knowledge than I have will have to comment, it seems like putting melodyne on and off a group of clips is using cakes render bit depth to make them into 1 big clip and replacing the original separate clips with this bounce wav ? Interesting experiment but way more than I understand what's up behind the scenes in melodyne or cake ? I haven't had the problem regularly that you describe of edits being frequently wonky, only happens for me if the edit is too far from the original, and I also never put melodyne on a bunch of clips at one time till this experiment. 

[marcL]

11 hours ago, treesha said:

you describe of edits being frequently wonky

I have made the same experience! Especially if you have melody jumps (more than a sixth) in the vocals or if you had to correct a wrong octave detection (in detection mode). In such cases the transition from one note to the next may sound strange or wonky when doing pitch/time changes.

Another problem with vocals and pitch detection in Melodyne is that it always assumes a 100 cent pitch value as correct! This is faulty for vocals (but correct for instruments)! Try it yourself, if you sing a melody with up and down jumps (maybe on the same syllable in the text), then it sounds only correct if the first note is on about 50 cents. You can also see this if you analyze vocal parts of well known artists. By the way, I learned this also in my music education.

[RexRed]

Yes, that is exactly my experience also Marled... When  notes jump a lot there is not enough bits thrown at them. It needs 2 pass variable bitrate, just like film clip rendering when there is an explosion or a lot of action in a scene, more bits get thrown at those sections... 

[RexRed]

Thank you Treesha for doing your testing of this problem. I have tried raising the khz but not raising the bitrate. I wonder if 32tbit recording might help this.

I wish I knew more about this problem and how to remedy it.

I think a variable bit recording might help this but how that would affect real-time monitoring while recording might be an issue.

I am not even sure if I can record in Cakewalk in 32bit my Steinberg UR22c is supposed to be 32 bit but I am not sure how that is applied and what about Cakewalk 64bit precision how does that fit in. And can Melodyne handle 32 bit recording? I am a bit overwhelmed by this and my song files would be a lot bigger I guess another 12tb hard drive might be in order. I will use the 12tb to store projects and the M.2 to record and edit projects (cloud storage is gonna love me). And with 32 bit recording my effects/CPU might get bogged down, maybe I'll upgrade my 12 core to 18 core. Here is where having my Nvidia 3090 assisting in the record/playback/rendering process would help. I wish I could  set the bitrate on a per track basis. Some tracks are background and some are up front. Just like in 3D art, background things can be low res and still be okay. Each track needs to be its own project layer container. Vocals up front and solo instruments would get the royal treatment.

Edited April 4, 2021 by RexRed

[RexRed]

Years ago I would have to program the drum machine and record that track first on my reel to reel before I would lay down any other tracks.

So every drum roll and cymbal crash had to be preplanned out before I would lay down any other instrument. Gone are those days.

Now I can just place a drum in any place in the song as I go along. What a change that was when technology developed enough to do that.

I seem to be at that same crossroad but now it is with the music and vocals.

I want to make the music lower resolution so the effects overhead is at a reasonable level.

And I want to make the vocal track at the very highest resolution possible. 

This would require that I make the music beforehand and then import it into a project with a higher bit and frequency.

Then lay down the vocals.

 I am wondering if there is a DAW out there that allows for different bitrates and frequencies  to run simultaneously.

Can Windows even do this and can ASIO and any Audio Interface do this? How about using something like VoiceMeeter Potato? 

I am thinking two instances of Cakewalk in sync to two PC's in sync... Sound cards slaving to the same clock or SMPTE?

Any ideas?

 

 

Edited April 4, 2021 by RexRed

[scook]

11 minutes ago, RexRed said:

I am wondering if there is a DAW out there that allows for different bitrates and frequencies  to run simultaneously.

While the DAW runs at one sample rate, CbB can run plug-ins at higher and lower sample rates

There are distortion and lo-fi effects that process audio at varying sample rates and bit depths.

[RexRed]

6 hours ago, scook said:

While the DAW runs at one sample rate, CbB can run plug-ins at higher and lower sample rates

There are distortion and lo-fi effects that process audio at varying sample rates and bit depths.

Thank SCook for bringing this awesome feature to my attention, I briefly heard about that once before and had not really thought about the implications.

It would be nice to be able to lower the sample rates and bit depths of certain tracks so I set the project to a very high rate and then decide as I go along which elements require that kind of precision. Maybe if these elements could have backup tracks so they resort back to the original rates when rendering. I am not so much concerned about project file size as I am about real-time playback CPU load as I like to mix and master simultaneously. 

[scook]

Start the project at a high sample rate.

The CbB upsample feature may be applied non-destructively on playback, during render or both.

I wrote a tool to make it easier to setup the plug-in settings.

It is one of the CbB tools mentioned in this thread

[RexRed]

9 hours ago, scook said:

Start the project at a high sample rate.

The CbB upsample feature may be applied non-destructively on playback, during render or both.

I wrote a tool to make it easier to setup the plug-in settings.

It is one of the CbB tools mentioned in this thread

I am going to try recording a project at 32bit with 48khz

I am assuming it will record the tracks at 32bits at 48khz. This logically means in my thinking that the 48hz will be subdivided up across 32bits.

I am assuming with more bits there will be much more resolution. I may need teo learn how to downsample the effects until the last minute before the final render.

I am hopping this will solve my problem with fuzzy transients in my Melodyne blobs.

The performance of the formants tool in melodyne is really poor and I am trying to fix this problem and I am hoping the throwing more bits at the problem will do it.   

I will let you know how this all turns out. I kept it at 48khz  because I think that audible range is enough.

I am excited to try out the downsampling effects capabilities in Cakewalk because I have a suspicion that at 32bit it will bog down my CPU even with 24 hyprethreads.

Projects capable of using varying sample rates would be a very nice feature and perhaps give Cakewalk yet again one more edge (along with many others) over other DAWS .

 

[msmcleod]

50 minutes ago, RexRed said:

I am going to try recording a project at 32bit with 48khz

I am assuming it will record the tracks at 32bits at 48khz. This logically means in my thinking that the 48hz will be subdivided up across 32bits.

I am assuming with more bits there will be much more resolution. I may need teo learn how to downsample the effects until the last minute before the final render.

I am hopping this will solve my problem with fuzzy transients in my Melodyne blobs.

The performance of the formants tool in melodyne is really poor and I am trying to fix this problem and I am hoping the throwing more bits at the problem will do it.   

I will let you know how this all turns out. I kept it at 48khz  because I think that audible range is enough.

I am excited to try out the downsampling effects capabilities in Cakewalk because I have a suspicion that at 32bit it will bog down my CPU even with 24 hyprethreads.

Projects capable of using varying sample rates would be a very nice feature and perhaps give Cakewalk yet again one more edge (along with many others) over other DAWS .

 

Bear in mind that currently most A/D converters only have a resolution of about 20 bits... they use oversampling to extend to 24 bits, but I doubt if you'll hear any improvement in quality going to 32 bit.  All you'll get by recording at 32bit is a bigger file.  To be honest, the only reason to have 32 bit or 64 bit files is if you're using an external audio editor, and you want them to use the extended dynamic range (although I guess in theory, Melodyne does come in to this category). 

You might get better quality going to 96Khz though.

BTW - you mentioned 64bit Double Precision Engine setting earlier in the thread... this has nothing to do with the bit size of your audio data. It only applies to the internal processing of audio data within Cakewalk's audio engine. In essence, it preserves more of the dynamic range when mixing a large number of tracks.

[RexRed]

2 hours ago, msmcleod said:

Bear in mind that currently most A/D converters only have a resolution of about 20 bits... they use oversampling to extend to 24 bits, but I doubt if you'll hear any improvement in quality going to 32 bit.  All you'll get by recording at 32bit is a bigger file.  To be honest, the only reason to have 32 bit or 64 bit files is if you're using an external audio editor, and you want them to use the extended dynamic range (although I guess in theory, Melodyne does come in to this category). 

You might get better quality going to 96Khz though.

BTW - you mentioned 64bit Double Precision Engine setting earlier in the thread... this has nothing to do with the bit size of your audio data. It only applies to the internal processing of audio data within Cakewalk's audio engine. In essence, it preserves more of the dynamic range when mixing a large number of tracks.

My Steinberg Audio Interface UR22C is 32bit/192khz  I think the problem has been the the same amount of bits get distributed across the frequency range. Higher bits get me a better editing resolution. 

I switched to 32 bit and last night notice the sound quality immediately in my first vocal take. The formants are still a bit fuzzy when edited in the extremes but on first observation I think they are about 50% better.

If I now switch to 96khz then that amount of bits will be distributed over a better frequency range. I am afraid though it might spread the bits thin again.

https://www.sweetwater.com/store/detail/UR22C--steinberg-ur22c-usb-audio-interface

Notice right under the product name it says 32bit.

Cakewalk handled this new project of 32bits perfectly and Melodyne opened the 32 bit vocal waves also.

I think I am really going to need to lower the resolution of effects if I got to 96khs I will come here and ask for assistance when I go to downsample my effects.

I don't think I need 96bit reverb.

This current project I am going to spread the 32bits over 48khz.

And then on the next project I will try moving up to 32bit 96khz.

It has not really been the frequency range, when left alone that frequency range is great. It has been the ability to edit those frequencies like pitch and formant where the character of the sound degrades considerably there are not enough bits to hold up the integrity and character of the waves when shifted.      

From gearslutz.com

The benefit of a full 32-bit playback path is greater CPU efficiency - to totally oversimplify, you're able to grab 32-bit packages of data from the DAW and pass them directly to the DAC chip without the CPU needing to rearrange them into 24-bit packages first.

The benefit of the 32-bit data stream in the DAC chip itself is improved filtering (resulting in lower noise and distortion), and superior-quality digital attenuation. 

Comment:

I have noticed my CPU meter bars barely budge, I thought they were not working at first. 

Edited April 5, 2021 by RexRed