What is happening to my voicebank?

[Daiko]

Hello everyone! I am currently working in a new voicebank for my friend.
He finished recordings and I immediately started to work on it.

However, there is something wrong with the bank.

The samples are normal, but there appears to be a problem with the FRQs OR WITH SOMETHING ELSE. Truth is: I don't know what's happening!
Could someone help me identifying the problem and the reason why most of the samples are sounding completely different from what he SHOULD SOUND LIKE, which in fact, can ALSO BE HEARD IN THE AUDIO, as the problem DOES NOT AFFECT ALL the sounds.

Please, give me a hand! Thank you.

AUDIO:

 

[Yomi]

How exactly do they sound different? Please explain the problem. Does it happen with all of the samples or only some of them? Because I've had a problem with some of my vbs before where some certain notes sounded strange (like growly) But I don't know if that's problem you are having or if it's the whole bank that sounds strange.

 

[Daiko]

How exactly do they sound different? Please explain the problem. Does it happen with all of the samples or only some of them? Because I've had a problem with some of my vbs before where some certain notes sounded strange (like growly) But I don't know if that's problem you are having or if it's the whole bank that sounds

Hello, thank you for stopping by! I just edited the thread with a soundcloud link so you can listen to an example.

 

[Yomi]

Hello, thank you for stopping by! I just edited the thread with a soundcloud link so you can listen to an example.

oh I hear it now! Hmm yeah it seems to be the similar thing that happened to my banks except your one I think more samples are affected. I feel like it COULD be a problem with the oto, but I'm not sure. If not, it could have something to do with how they were recorded, but if those samples sound fine, then I'm not sure what it could be other than the oto. But eh I could be wrong though ^^ I hope you'll be able to fix it!

 

[Kiyoteru]

This could be an FRQ error where the pitch was estimated to be at the wrong octave. Press this button to edit the FRQ for the selected note.


It's a bit difficult to see, but the buttons in the bottom left corner are x3, x2, delete, /2, /3. You can select all of the affected points, and multiply/divide the frequency values by double clicking the buttons as necessary.

 

[Daiko]

This could be an FRQ error where the pitch was estimated to be at the wrong octave. Press this button to edit the FRQ for the selected note.
View attachment 10309

It's a bit difficult to see, but the buttons in the bottom left corner are x3, x2, delete, /2, /3. You can select all of the affected points, and multiply/divide the frequency values by double clicking the buttons as necessary.
View attachment 10310

Hello, Kiyoteru-sensei! Thank you once again for your help and care.

I will try fixing the FRQs manually. After doing so, I'll post a new audio in this same thread, so we can check the results and get to a conclusion, is that ok?

Truth is I've tried fixing NOAH's FRQs already, in a plugin called 'frq editor': once, when one of my latest VB's pitches was acting the same way, I asked a friend on how to fix it and she just did it by herself wtih frq editor. I've been trying to replicate what she did with NOAH's case, but have not made it with yet... I must understand what she did so I can fix this. I tried speaking to her but she couldn't make me understand...

Anyway, I'll be fixing some of the frqs for now, in order to do a new example and I'll post it here and we'll see what happens.

 

[Daiko]

This could be an FRQ error where the pitch was estimated to be at the wrong octave. Press this button to edit the FRQ for the selected note.
View attachment 10309

It's a bit difficult to see, but the buttons in the bottom left corner are x3, x2, delete, /2, /3. You can select all of the affected points, and multiply/divide the frequency values by double clicking the buttons as necessary.
View attachment 10310

Hello Kiyoteru-sensei, I'm back.

I've just adjusted enough FRQs to render the following phrase. Still, the issue persists: the same massive difference in quality with no aparent reason... Could it really be the samples? But if that was the case, why are there samples that are sounding just fine? They were all recorded the same way...

Audio:

 

[nneko]

My suggestions are:
1. Delete the frq files and generate new ones.
2. Make sure your samples are mono instead of stereo; I've had issues with stereo samples before.
3. Check the sample rate of your files (I don't know if sample rates affect anything, but who knows?).

 

[Daiko]

Oto is normal, all .wav samples are mono.

All of the banks' samples have the exact same audio quality.

Even as frq are corrected, sound quality disparity between working and mysteriously non-working sounds does not end.


Example of "Bad sound"



Example of "Working perfectly fine sound"


Am I missing something...?

 

[nneko]

Have you changed resamplers or wavtools recently?
Does the ust contain any flags that might be conflicting with your voicebank?
Are there plugins that may cause an issue?

This is all I can currently think of at the moment, since I can't troubleshoot the problem directly.
I hope you can find the solution soon, because it sucks putting all that work into a project and having it not work out.

 

[Daiko]

Have you changed resamplers or wavtools recently?
Does the ust contain any flags that might be conflicting with your voicebank?
Are there plugins that may cause an issue?

This is all I can currently think of at the moment, since I can't troubleshoot the problem directly.
I hope you can find the solution soon, because it sucks putting all that work into a project and having it not work out.

I'm using wavtool4vcv and moresampler. I've also used moresampler as both wavtool and resampler.

I've also used TIPS, fresamp11, fresamp014omp and even dll to listen to how it sounds.. The inconsistencies persist with all mentioned resamplers.

 

[nneko]

In the "bad sound" image, it looks like UTAU is having trouble generating the frequency, and so some parts are lost.
Sometimes you can fix it by editing the frq and adding the missing parts, but that can also cause other problems.

I wonder if there is some subtle noise in the samples that the resamplers don't like? Like too much air going into the mic, but without it popping.
I would try re-recording one of the bad samples as a "version 2" and then compare the frq between version 1 and version 2.

(Minor edits.)
(Edit 2: The one time I needed my samples to be broken and UTAU's like "nah, they're good when I say they are." I've omitted my failed test.)

 

[Kiyoteru]

Did you record at a pitch near C3 or C2?
If you recorded at C2, then the FRQs that say C3 are wrong. Select all the frequency points and use the /2 (divide by 2) button.
If you recorded at C3, then the FRQs that say C2 are wrong. Select all the frequency points and use the x2 (multiply by 2) button.

 

[Daiko]

Did you record at a pitch near C3 or C2?
If you recorded at C2, then the FRQs that say C3 are wrong. Select all the frequency points and use the /2 (divide by 2) button.
If you recorded at C3, then the FRQs that say C2 are wrong. Select all the frequency points and use the x2 (multiply by 2) button.

It was recorded at C3.

I just fixed some of the awful sounding ones by getting their frqs to the third octave and they haven't changed.

Is there any possible reason other than FRQs for such an immense quality disparity between the normal and wrecked sounds, when all the samples sound the same? This is so intriguing...

 

[Arytii]

can you give some of the sampless that had problems in utau? maybe there's something interfearing with the samples?
i had problems like this too like a fan was on behind me the whole time and i had to rerecord the whole voicebank ;-;

 

[Daiko]

can you give some of the sampless that had problems in utau? maybe there's something interfearing with the samples?
i had problems like this too like a fan was on behind me the whole time and i had to rerecord the whole voicebank ;-;

Hello, Arytii, thak you for stopping by! I made this new audio so we can take a look at what could be going on with my friend's recordings and how they've been read by UTAU.

Audio:

I can somewhat listen to higher noise present in the faulty ones, but is that REALLY what's causing this massive difference? I've NEVER seen this happen...
I'm afraid he won't be able to re-record, too...