Instrument UTAU voicebank: Need help and feedback!

[Mougeki Mero]

Hello guys! I am new here, but this isn't the introduction board, so I won't take too long XD
So, I am developing an UTAU voicebank, specifically, an Acoustic Guitar VB. I have already around 260 samples recorded, an edited model (with some of my own modelling) and official illustrations going on(thanks to some japanese friends).

So, I wanted feedback and advice about it. Here is the "concept art" I made:http://ashley-andred.deviantart.com/art/UTAU-Concept-Art-Mougeki-Mero-590524493

Here is the edited model: http://ashley-andred.deviantart.com/art/MMD-WIP-Mougeki-Mero-UTAU-592297792

Here is a sample:

So, what I need help with is...the best way to play the guitar in UTAU. The samples are all from real acoustic guitar (from a company that allowed me using them) and from kinda HQ guitar. The problem is that inside UTAU it sound a bit distorted. I checked my illustrator friend's UTAU (http://amata.sonnabakana.com/sohobass.html) and it sounds very clear and real like an eletric bass should be. However, unlike Soho, my Acoustic guitar VB sounds a bit distorted when on UTAU. I figured it sounds more real using C99W99 flags and with resampler or TIPS or even better with moresampler (with no flags). Anyways, it still sounds kinda distorted. So, any ideas to help me? XDDD Thank you guys!

PS. Send me feedback too please XD Thanks again

 

[Kiyoteru]

Have you tried deleting pitchbends and MOD? I believe that pitchbends are occassionally useful for recreating guitar, but not so much as a transition between every note, like vocals. Removing MOD should allow the samples' natural variation in pitch come through.

(btw, being that UTAU is meant for vocals, I'm sure you'd have much better results putting your guitar samples into a format meant for instruments... Perhaps a soundfont, or something like DirectWave or Kontakt.)

 

[Mougeki Mero]

Have you tried deleting pitchbends and MOD? I believe that pitchbends are occassionally useful for recreating guitar, but not so much as a transition between every note, like vocals. Removing MOD should allow the samples' natural variation in pitch come through.

(btw, being that UTAU is meant for vocals, I'm sure you'd have much better results putting your guitar samples into a format meant for instruments... Perhaps a soundfont, or something like DirectWave or Kontakt.)

Thanks for the tips!!!! I have deleted the MOD but not the pitchbends. I have to try this after work. Oh, about why UTAU...XD well, because I have alwayd wanted to have an UTAU VB, but my microphone is bad. Cant buy a better one. And also, I got inspiration from Soho's VB, so I wanted to have instrument VB as well.


Thanks again!

 

[Kanru Hua]

IMHO, obviously guitar is an instrument fundamentally different from speech, so you can't expect a speech modification algorithm to work well on non-speech audio signals. The following are specific reasons from a technical point of view.

• Both Moresampler and World are based upon the minimum-phase property of vocal tract so for these resamplers (and their derived works) phase property of guitar sound would totally lost (precisely speaking, the phase loss for Moresampler maybe smaller than World).
• Currently all resamplers assume the harmonicity of speech. This assumption is not so bad for speech, but is bad for guitar sound whose "harmonics" slightly deviate from integer multiples of fundamental at high frequencies. The degree of inharmonicity actually depends on the elasticity of string used in that guitar.
• Moresampler interpolates LLSM parameters during phoneme transition, unlike most other wavtools which cross-fade the waveform. This is nice for speech because it is analogous to moving your organs during a continuous phonation, instead of having two people singing at the same time with cross-fading volume. But, guitar strings work in the opposite way. When you pluck two strings, obviously they vibrate at the same time.
• The most important point: guitar doesn't preserve spectral envelope during pitch shifting.

If you still want to synthesize guitar in UTAU, I'd suggest picking a resampler whose algorithm isn't too bad for guitar. For example, the default resampler or fresamp with "formant filter" turned off.

 

[Mougeki Mero]

IMHO, obviously guitar is an instrument fundamentally different from speech, so you can't expect a speech modification algorithm to work well on non-speech audio signals. The following are specific reasons from a technical point of view.

• Both Moresampler and World are based upon the minimum-phase property of vocal tract so for these resamplers (and their derived works) phase property of guitar sound would totally lost (precisely speaking, the phase loss for Moresampler maybe smaller than World).
• Currently all resamplers assume the harmonicity of speech. This assumption is not so bad for speech, but is bad for guitar sound whose "harmonics" slightly deviate from integer multiples of fundamental at high frequencies. The degree of inharmonicity actually depends on the elasticity of string used in that guitar.
• Moresampler interpolates LLSM parameters during phoneme transition, unlike most other wavtools which cross-fade the waveform. This is nice for speech because it is analogous to moving your organs during a continuous phonation, instead of having two people singing at the same time with cross-fading volume. But, guitar strings work in the opposite way. When you pluck two strings, obviously they vibrate at the same time.
• The most important point: guitar doesn't preserve spectral envelope during pitch shifting.

If you still want to synthesize guitar in UTAU, I'd suggest picking a resampler whose algorithm isn't too bad for guitar. For example, the default resampler or fresamp with "formant filter" turned off.

Thanks for the tips!! I know it is hard to, but even so, what I dont get is how other guitar used in UTAU sounds good while mine looks distorted (still it is small distortion, but it is not like i want it to be). I can get it very similar to real when i put this settings:

Resampler: C99W99F0B0Y0H0
Modulation: 99% sounds good IDK why
Mode 2
No pitching

EDIT: MANAGED TO FIX IT, NOW SHE SOUNDS REALISTIC, THANKS FOR ALL THE REPLIES!!!