How do you record a whisper voicebank?

Mochlin

Ruko's Ruffians
Defender of Defoko
Something like Mawarine Shuu's CVVC recordings, whenever i record like that OREMO doesn't detect any sounds and it doesn't tell me the note.
 

Nohkara

Pronouns: He/him
Supporter
Defender of Defoko
People often call VB that is recorded with a soft airy voice that is not actual whispering as "Whisper" VB. There are not many Whisper VB that are actually recorded with REAL whispering samples because UTAU usually doesn't work with these sounds. Often people tell that "voicebank with REAL whispered voice doesn't work" but there are tricks to make it work #ihavemadeone

I recommend that you record it as CV or as 2-mora EVE CVVC. Unlike with regular VB, use Audacity instead. I highly advise having a pop filter when recording a real whispering because many consonant sounds will more likely pop out than usual.

If you record as CV, I recommend recording more than one syllable per wav file (because CV, you can take a break and breathe between syllables like I did) e.g. "ka+ki+ku+ke+ko.wav". If as CVVC, something like "kaka+kiki+kuku+keke+koko+nka.wav" where + means a break), I'll explain more why.

Step 1) Open Audacity

Step 2) Record a string (e.g. "ka+ki+ku+ke+ko.wav" or "kaka+kiki+kuku+keke+koko+nka.wav" whatever you prefer)

Step 3) Select all and choose "normalize" from Effects. BOOM! Your real whispering samples are louder. But to make absolutely sure that these samples run smoothly in UTAU without crashing down, let's make another little trick.

Step 4) Add a super short "a" sound from your normal VB recordings the end of the string. This super short "a" will make sure that UTAU is able to generate FRQ file for the wav file - this is a common reason why normal whisper VB doesn't work in UTAU because UTAU is not used to generate FRQ files for real whisper samples or breathes.

So at the end, it should look less or more like this:

How to real whisper vb.png

Step 5) From Files, "Export Audio..." and export is as "WAV 16 bit" format! Don't use any other bit formats than 16, 16-bit WAV works in UTAU only!

Step 6) Now you have saved it, let's reply step 2 with the next sample.

The reasons why I said that more than one string is mainly because that will be much faster! You probably prefer to do this up to 50 times than 160-200 times, don't you?

And here is a sample how my real Whisper CV VB sounds like that is done with this technique:



Good luck ^^
 

Sors

Local Guppie & UTAU Korean Advocate
Tutor
Defender of Defoko
Something like Mawarine Shuu's CVVC recordings, whenever i record like that OREMO doesn't detect any sounds and it doesn't tell me the note.
Do NOT record it like Shuu. His whisper samples usually get messed up in UTAU. Instead, record soft, airy and higher than your normal base pitch. When we whisper, we tend to be higher.
 
  • Like
Reactions: Mochlin and partial

Nohkara

Pronouns: He/him
Supporter
Defender of Defoko
Thanks :3 i'll try to record it like that, your voicebank is really nice.
Also, what resampler did you use? Or what resampler would you recommend
Thank you! I'm glad that you find my reply helpful! ^^

Since I use UTAU-Synth (Mac UTAU) only, I don’t/cannot use resamplers... I don't know what resampler(s) works best, so you'll need to test out by yourself.

Oh, and my last tip: Super commonly real whispering and screamo voices don't like long notes and the vowels stretch out very unnatural sounding. To fix that "slice" a long note to shorter notes like this:

Screen Shot 2017-11-03 at 18.14.05.png
 

Similar threads