Help me with recording and explain frqs?

P

partial

Guest
OK, so this thread is probably odd coming from me.

Let me start off by saying, I cannot edit frqs because I don't know how to do it properly, so I always record my vbs very quietly. I found an old vccv of mine from 2015, with the volume of the wavs at a more average level, and the voice that bank creates is much less robotic and much more natural. It's quite glaring and very problematic to me that this bank I've been working hard on sounds worse than my first voicebank.
Here's a sample comparing that 2015 vccv to the vccv I've been working on. Rendered with moresampler for both banks.


Here is a comparison of a few oto configs, so you can see the difference in volume.
upload_2017-12-5_6-44-48.png
upload_2017-12-5_6-44-23.png
upload_2017-12-5_6-47-56.png
upload_2017-12-5_6-48-35.png
upload_2017-12-5_6-50-54.png
upload_2017-12-5_6-50-18.png
You get the idea.
The reason I started recording as I do now is because I don't run into frq problems, but I just realized the quality I've sacrificed.

Does anyone understand how to use this frq editor? I can't use the editor built into UTAU, as the buttons at the bottom get cut off by the edge of the window.
upload_2017-12-5_6-52-37.png
 

VocAddict

The Voice Within Us
Defender of Defoko
I'm on my phone so I may leave out some parts, might add in the rest later.

Also, I think the button at the bottom gets cut off for all of us lol.

It's pretty simple.

You load a bank (or a sample) and you get the window on the right. You already know how to access the spectrogram by clicking the 'S' and the '*' next to it would affect the intensity of the frequncies shown.

The red line shows the average pitch for all of the samples of the bank while the blue boxes that are within the sample itself show the pitch of it.

If you scroll down on the left, you'll end up seeing the frequency numbers for each of those blue boxes in the sample. To fix a pitch that may be off, you may either click the cell and edit it with the update button at the top, or you can manually drag the box to where you want it. The second option is far more easier lol.

In the tool menu, if I remember correctly, you can set it to find errors in your banks and it will give you suggestions on how to fix them (automatic) though some of them aren't really good so you'll have to do them by hand. What you want to do is have most, if not all of the blue boxes on the red line.

Now, if the bank you're doing is multipitch, load each pitch by itself otherwise you'll end up with a horrible sounding bank by the time you're done with it.

frqeditor allows you to play back the sample after you've edited it, either directly (without the use of the resampler) or with a resampler. You can set which resampler you want in the tool menu as well, set any flags and modulation, and click play.

There should be another window with it, it'll show the list of the samples and the frequency map type (so, for resampler and fresamp if it's your selected resampler, you'll see .frq, for tips .pmk, and so on) and it'll show if you've edited any of those. You can discard those changes by reloading that individual samples (or the bank if that's what you want), or save that sample by clicking save in the file menu, or save all for the entire bank.

If there's any missing frequency files, you can have your selected resampler make them from the tool menu as well, edit where needed.

So, for the picture you've posted at the end, There isn't much discrepancies between the pitches so you can leave it as is. You shouldn't really edit the frq files if nothing is wrong. Since this is a VV sample, this is the normal frq map.

Louder samples makes it easier for resamplers to make frq maps for them since most of the noise (both from the mic and from the environment) gets drowned out.

That's why with softer ones, you tend to get errors since the resampler would start the frq noise since it's closer the it in volume, and sometimes, they may not do the frq map correctly and completely ignore parts of the sample. So, it's important to go thorugh and deal with those manually. So, softer banks tend to end up with more noise in their samples when it comes to rendering and it gives off that unnatural, robotic feeling since UTAU has to raise the volume of the sample completely to render it.

And if you think your latest bank is recorded softly, take a look at my Arpasing bank lol. So you can imagine the amount of noise that ends up in the rendering process.

upload_2017-12-5_10-6-14.png

I think that's it. I hope this helps. I probably missed a few things but trial and error all the way!
 
P

partial

Guest
Thread starter
Also, I think the button at the bottom gets cut off for all of us lol.
Well that's a relief! I thought I was the only one and I hated not knowing for sure what I was clicking on!
frqeditor allows you to play back the sample after you've edited it, either directly (without the use of the resampler) or with a resampler. You can set which resampler you want in the tool menu as well, set any flags and modulation, and click play.
I did not know it played the edited sample since I haven't played anything with the button, this is useful to know!
Now, if the bank you're doing is multipitch, load each pitch by itself otherwise you'll end up with a horrible sounding bank by the time you're done with it.
I am so glad you mentioned this, because my bank is multipitch and I would be so upset if I ruined it. (Though at this point, I'm feeling the need to scrap it and start over...) I don't really know why you'd need to load each pitch separately, though. Is it because when you load all pitches at once, the average frequency of the bank sort of levels out? If that makes sense. (when you click on the UTAU character's icon and it shows the "voices", etc.)
Louder samples makes it easier for resamplers to make frq maps for them since most of the noise (both from the mic and from the environment) gets drowned out.
I didn't know this, either. I always assumed UTAU had it easier when making frq files for quiet samples. I never get strange jumps or anything when I record, but I record fairly quiet. Generally when I record as I did for me 2015 VCCV, I get hiccups and jumps quite often.
And if you think your latest bank is recorded softly, take a look at my Arpasing bank lol. So you can imagine the amount of noise that ends up in the rendering process.
That's so quiet, how did you even manage to oto that?! I'm sure it was pure spectrogram, right?

Thanks for your reply! It's been very useful!!
 
  • Like
Reactions: VocAddict

WinterdrivE

Ritsu's Renegades
Defender of Defoko
frq's are a moot point if you're using moresampler because moresampler doesn't use frq's.

VocAddict pretty much runs you through how to edit the frq's though. Typically if something's wrong with them, its that it miscalculated which octave some of the points are in. Pitches that're an octave apart are twice/half the frequency of each other. eg if A4 is 440 Hz, A1 is 55, A2 is 110, A3 is 220, A5 is 880, A6 is 1760, and so on. So as you look at the frequency number in the lext box in the frq editor, if you see anything that's wildly higher/lower than what's around it, look closely and chances are they're around double or half of what surrounds them. If so, select them and with the x2 button (if they're lower) or /2 button (if they're higher)

[edit]
And about the quality thing, that's like also because of what VocAddict said. When you record more softly, your microphone's noise and any ambient noise are that much louder compared to the actual vocals.

And also hahahahaha when i'm recording higher pitches I have to back off the mic to keep them from clipping (and they still do OTL)
 
Last edited:
P

partial

Guest
Thread starter
frq's are a moot point if you're using moresampler because moresampler doesn't use frq's.

VocAddict pretty much runs you through how to edit the frq's though. Typically if something's wrong with them, its that it miscalculated which octave some of the points are in. Pitches that're an octave apart are twice/half the frequency of each other. eg if A4 is 440 Hz, A1 is 55, A2 is 110, A3 is 220, A5 is 880, A6 is 1760, and so on. So as you look at the frequency number in the lext box in the frq editor, if you see anything that's wildly higher/lower than what's around it, look closely and chances are they're around double or half of what surrounds them. If so, select them and with the x2 button (if they're lower) or /2 button (if they're higher)
I've just been using the avg button. Is that not what I'm supposed to do? I get frustrated with the x2/x3 and /2 and /3 buttons because they generally don't end up putting the messed up sections where I feel it should be.
 

WinterdrivE

Ritsu's Renegades
Defender of Defoko
I mean, as long as it sounds OK, then you're fine. I haven't messed with frw's in a long time cuz I've pretty much only been using moresampler, but I'd be hesitant to use the average button only because it just creates a straight line through those points. And again, as long as it gets it close enough and it sounds fine, its better than the frq being totally messed up, but I'd be concerned about that random flat line not as accurately representing the pitch as possible. eg, what if there was a slight bump in the pitch along those points that isn't corrected properly in the frw because there's just a flat line going through it with the avg button or something like that. But if the x2 or /2 buttons aren't fixing it right, then the avg button is really your only other option so its definitely better than not fixing it at all.
 

VocAddict

The Voice Within Us
Defender of Defoko
frq's are a moot point if you're using moresampler because moresampler doesn't use frq's.
I believe frqeditor has support for moresampler .llsm and .mrq files iirc. moresampler's frequency map is in the desc.mrq file so that's what you edit in frqeditor

That's so quiet, how did you even manage to oto that?! I'm sure it was pure spectrogram, right?
Pure spectrogram lol. I didn't know what volume was when I was otoing that bank and fixing errors.
 
  • Like
Reactions: kimchi-tan
P

partial

Guest
Thread starter
I mean, as long as it sounds OK, then you're fine. I haven't messed with frw's in a long time cuz I've pretty much only been using moresampler, but I'd be hesitant to use the average button only because it just creates a straight line through those points. And again, as long as it gets it close enough and it sounds fine, its better than the frq being totally messed up, but I'd be concerned about that random flat line not as accurately representing the pitch as possible. eg, what if there was a slight bump in the pitch along those points that isn't corrected properly in the frw because there's just a flat line going through it with the avg button or something like that. But if the x2 or /2 buttons aren't fixing it right, then the avg button is really your only other option so its definitely better than not fixing it at all.
I can also just draw new lines with my tablet, but I've got a bit of a shaky hand, so it never looks as good as I'd like. I don't use the avg button for the whole frq, just to correct strange spikes. I try to get it close to the red line. That's the goal, right?
 

WinterdrivE

Ritsu's Renegades
Defender of Defoko
Not necessarily. If the frq was completely correct yes, and if its not horribly messed up, also yes (for the most part). But the red line, as VocAddict also mentioned, is the average frequency of the frq. However, its the average frequency of the frq as it currently is. eg, I have a sample that's recorded on D4, but a good 1/3 of the frq was wrongly detected an octave higher, so the red line is sitting at F4 because all of those wrong points that're at ~570 Hz instead of ~285 are bringing the average up.

Here's a picture
Kl5QtaA.png


The blue text and line is where the frq points are supposed to be (around D4), the dark red line is where the wrong ones are (around D5). Note how the red average line is on around F4 because all those wrong points on D5 are pulling it up. Also notice the numbers on the side I have circled (and highlighted, but the red highlight is hard to see on the right side). Note that they're about 2x the numbers around them, confirmed by the right side where they're an octave higher, on D5 instead of D4. Using the /2 button halves the numbers and brings them back down to D4, where they should be. This retains the frq's original detection of the pitch curve (any slight bumps or changes in pitch, since humans aren't really capable of holding a perfectly steady pitch) while moving the points back down to where they should be.

Again, idk that the issue with you is necessarily that the octave is wrong. (that's the most frq error I've encountered) But if it is, this is how I'd recommend fixing it, though.

[edit]
Also, i just noticed that you said you used two different mics between the two VBs, and that could have a big impact on the quality difference between them.
 
Last edited:
P

partial

Guest
Thread starter
I see.

Generally, my frqs on or very very close to the red line. So what you're saying is it's not really about the red line, just that the blue dots are relatively in the same place?
upload_2017-12-5_17-4-9.png
upload_2017-12-5_17-4-47.png
upload_2017-12-5_17-5-26.png
 

WinterdrivE

Ritsu's Renegades
Defender of Defoko
I see.

Generally, my frqs on or very very close to the red line. So what you're saying is it's not really about the red line, just that the blue dots are relatively in the same place?
Yeah, pretty much. I don't see anything obviously wrong with the screenshots you posted, though. Unless you hear something wrong, you can probably leave them alone.
 
P

partial

Guest
Thread starter
Yeah, pretty much. I don't see anything obviously wrong with the screenshots you posted, though. Unless you hear something wrong, you can probably leave them alone.
Ah, nah. Those were only posted as examples of where the blue dots are compared to the red line in most of my samples.

I wanted to rerecord English, but louder like the old bank. That's when I would have problems, though.
 

WinterdrivE

Ritsu's Renegades
Defender of Defoko
Ah ok.

The only other thing I want to add is that any sudden changes in the frq right around the consonants are probably fine. eg, the pitch jumps up slightly at the beginning of some of those syllables in the screenshots you sent. I wouldn't worry about things like that unless it actually sounds super wrong once its run through the resampler. Around consonants, the vocal cords may relax causing the pitch to jump up or down immediately before, during, and immediately after the consonant, so the frq may not be 100% flat, and that's fine.

If you do run into problems when you get to it, i'd be curious to see what's going on in the frq.
 
P

partial

Guest
Thread starter
Ah ok.

The only other thing I want to add is that any sudden changes in the frq right around the consonants are probably fine. eg, the pitch jumps up slightly at the beginning of some of those syllables in the screenshots you sent. I wouldn't worry about things like that unless it actually sounds super wrong once its run through the resampler. Around consonants, the vocal cords may relax causing the pitch to jump up or down immediately before, during, and immediately after the consonant, so the frq may not be 100% flat, and that's fine.

If you do run into problems when you get to it, i'd be curious to see what's going on in the frq.
Ok! I'll definitely post here once I get on redoing the multipitch bank I've already started on. I think the current will be scrapped...
 
  • Like
Reactions: WinterdrivE

Similar threads