doppeltler & f2resamp - Ameya's new WIP resamplers

na4a4a

Outwardly Opinionated and Harshly Critical
Supporter
Defender of Defoko
Surprised no one has made a thread about this!

Ameya has been slowly working on UTAU related updates, fairly recently they have been pushing out their work in progress resampler, doppeltler. It's a completely new resampler, fully multithreaded, and has 64-bit support.

Supports 16, 24 , and 32 bit wave files.
Supports Mp3, OGG Vorbis, and FLAC encoded voicebanks as long as the decoders are set in your PATH system variable. These will be represented in the oto.ini as ".mp3.wav".

The full readmes and addition information are available at http://utau2008.xrea.jp/2020/engine/

Recently Ameya has also released yet another engine, f2resamp. f2resamp is part of the fresamp family and is a complete, from scratch, recode.




Supported flags:

Standard Flags:
  • g - gender (same as resampler)
  • B - Breathiness (same as resampler) [Default B50]
  • N - No Formant Correction (same as resampler)
  • P - Loudness Normalization (same as resampler) [Default P80]
  • G - recreate FRQ files

New Flags:
  • a - "High frequency emphasis rate of the noise component" [Default a100]
    • Think of this as high frequency breathiness, while B is low frequency.
    • Has range of a0 to a1000
      • a0 - a100: ??? (reduces existing frequencies, but Ameya claims the function is different)
      • a101 - a1000: boosts high frequencies
    • Does not remove voiced component the same as B100.
    • Can be used in conjunction with B
  • R - Reverse sample, without reversing waveform. (Tweet)
  • S - Thread count (uses all available cpu threads by default)
    • If you CPU has 4 cores, the default value will be S4.
  • i - "length of the rear fixed area (ms)"
  • v - "consonant velocity of the rear fixed area"
v0.05+:​
  • t - fine pitch adjustment in cents
    • range of 199 to -199
  • n - Silence mode (does nothing as of 0.05)
v0.08+​
  • t - fine pitch adjustment in cents (replaces T)
    • range of 199 to -199
As of now, Doppeltler and F2resamp have the same flag support. With some minor changes.
  • a - Same as Doppelter, default is a50 instead of a100.




Downloads (Links to tweets):
0.01 (no longer available, have archived personally.)
0.02 (no longer available, have archived personally.)
0.03 (no longer available, have archived personally.)
0.04
0.05
0.06
0.07
0.08


You may also download a 64-bit version of the default wavtool at http://utau2008.xrea.jp/2020/engine/

Ameya has also release a complete rework of fresamp, called f2resamp

Download (links to tweets)
0.01
0.02
0.03

Be sure to check their Twitter for more updates, it's all in Japanese but machine translation is good enough to get the gist.



Notes:

Some things to consider when using these resamplers:

Doppelter:
- Version prior to 0.04 generate flawed FRQ files. You should generate with a different engine or use 0.04 to generate FRQs
-If a note is too short you may get unexpected results due to how doppeltler handles otos. If you find end-breaths being cut short then extend the length of the breath's note.
- It's an unfinished resampler, expect problems. If it's a persistent issue across multiple banks and you can reproduce it then you could probably report it.
- Versions before v0.07 have some issues regarding FRQ generation, use a later version if you encounter it.

F2resamp:
- F2resamp is very sensitive to FRQ file flaws, bad FRQ files can easily affect the render result. One example is the lack or alteration of unvoiced fricative and sibilant consonants due to the estimated pitch increasing unexpectedly.
- Version 0.02 has debug code that may cause the file to be unreadable by some DAWs, 0.03 fixes this bug and is otherwise the same version.
 
Last edited:

na4a4a

Outwardly Opinionated and Harshly Critical
Supporter
Defender of Defoko
Thread starter
Ameya just released v0.05. This adds the T and n flags. Added link in first post.
 

na4a4a

Outwardly Opinionated and Harshly Critical
Supporter
Defender of Defoko
Thread starter
Ameya released v0.06, link added.
 

na4a4a

Outwardly Opinionated and Harshly Critical
Supporter
Defender of Defoko
Thread starter
v0.07 and v0.08 is out, links added.
v0.07 is more or less the same as v0.06 but with imporoved frq generation.
v0.08 changed the T flag to t.
 

na4a4a

Outwardly Opinionated and Harshly Critical
Supporter
Defender of Defoko
Thread starter
Ameya has released f2resamp, it appears to be a fresamp update/recode.

You can download it here, it is also added to the first post.
 
Last edited:
  • Like
Reactions: VocAddict

na4a4a

Outwardly Opinionated and Harshly Critical
Supporter
Defender of Defoko
Thread starter
Ameya has updated f2resamp to 0.03.
0.02 and 0.03 are the same, 0.03 just has some debug code removed.
This version fixes a bug in multi-threading and appears to increase max pitch estimation.
There is a bug where the lack of a right cutoff/end region will crash the engine.

Links added to first post.
 
Last edited:

Similar threads