Repository navigation
Installation
Thanks for taking the time to read Tiger's usage guide! Below you will find some information to assist in using Tiger's DiffSinger models.
Download the voice library here!
To install Tiger, simply drag and drop the TIGER_DS_b08.zip file into OpenUTAU. Accept any windows that open up. TIGER_DS_b08.zip contains everything you need to get started!
- The current
dsdictis not perfectly optimized, so you may need to do some phonetic editing. -
tranddrare lacking data in the voice bank overall, so you may still get better results usingch randjh rrespectively.
This voicebank is meant to be used in OpenUTAU, and should be propely configured upon using for the first time. However, there are a few different options to choose from to change the way you use Tiger!
I'm still waiting for DIFFS EN X to be implemented into OpenUTAU, so until then, use DIFFS EN for English. The included dsdict.yaml accounts for many instances of where the DIFFS EN X would alter pronunciation.
By default, Tiger will produce the best results with the DIFFS EN X phonemizer, which is a phonemizer I contributed to OpenUTAU that utilizes the [ax] and [dx] phonemes for more accurate American English. This phonemizer reads from dsdict-enx.yaml in the main folder, as well as in dsdur and dspitch.
To overwrite the lyrics typed in each note, you can add [] after the lyrics and include phonemes between the brackets. For example: cut[k ax t].
The DIFFS EN phonemizer is supported, but not by Tiger officially, and will simply use the estimated output from the ArpabetG2p. (Not Recommended)
If you would like to do phonetic only input (bypass the [] requirement), you can use the basic DIFFS phonemizer. An added bonus to this mode is the ability to read a phoneme table for Japanese Hiragana from dsdict.yaml.
By default, OpenUTAU will configure your preferences to render DiffSinger voices on CPU, but I recommend switching to GPU rendering to optimize speed and make using OpenUTAU a more enjoyable experience. To do so, in the preferences menu change the following:
Rendering:
Machine Learning Runner: directml
GPU: Your GPU, for example mine is "[0] NVIDIA GeForce RTX 3070"
DiffSinger Render Speedup: 50 for usage, 5-10 for rendering.
DiffSinger Render Depth: 400Tiger can use 5 different voice modes currently, here's a quick rundown of each. This page has more information on each voice mode.
| Voice Mode | Description |
|---|---|
| ☮️ Fresh ☮️ | is the basic vocal mode. |
| 💃 Disco 💃 | is a softer, calmer vocal mode. |
| 🌩️ Electric 🌩️ | is a strong, shocking vocal mode. |
| 💿 Vinyl 💿 | is a mature, booming vocal mode. |
| ❓ Mystic ❓ | is a very soft, whispery vocal mode. |
| 🌺 Falsetto 🌺 | is a vocal mode focused on high range falsetto. |
| ✨ Cetera ✨ | contains data for other languages, such as Japanese and Spanish. Closest to Fresh in tone. |
If you run into any issues using Tiger, reach out to me here via the issues tab, or by other means such as twitter or discord.
🗣️ Phonetic Information 🗣️
English Specific
| Vowel | Explaination | Consonant | Explaination |
|---|---|---|---|
| aa | body [b aa d iy] | b | bubble [b ah b ax l] |
| ae | cat [k ae t] | ch | check [ch eh k] |
| ah | cut [k ah t] | d | duty [d uw dx iy] |
| ao1 | all [ao l] | dx | butter [b ah dx er] |
| aw | out [aw t] | f | fish [f ih sh] |
| ax2 | again [ax g eh n] | g | glob [g l aa b] |
| ay | fly [f l ay | hh | help [hh eh l p] |
| eh | get [g eh t] | jh | joke [jh ow k] |
| er | her [hh er] | k | calf [k ae f] |
| ey | play [p l ey] | l | little [l ih dx ax l] |
| ih | been [b ih n | m | make [m ey k] |
| iy | bee [b iy] | n | noodle [n uw dx ax l] |
| ow | crow [k r ow] | ng | thing [th ih ng] |
| oy | boy [b oy] | p | price [p r ay s] |
| uh | good [g uh d] | r | rice [r ay s] |
| uw | too [t uw] | s | slay [s l ey] |
| sh | ship [sh ih p] | ||
| t | tip [t ih p] | ||
| th | thigh [th ay] | ||
| v | vibe [v ay b] | ||
| w | what [w ah t] | ||
| y | yes [y eh s] | ||
| z | zoo [z uw] | ||
| zh | asia [ey zh ah] | ||
| tr | truck [tr ah k] | ||
| dr | dress [dr eh s] |
1: [ao] is a multiuse phoneme dependant on context. Preceeding [l], it makes "all", but preceeding [r], it makes "or".
2: [ax] is a schwa, which is another multiuse phoneme dependant on context. It creates [el] when it preceeds [ax l], for example.
Usage Basics
| Phoneme | Name | Explaination |
|---|---|---|
| SP | Silence | DiffSinger's label for silence |
| AP | Breath | DiffSinger's label for breaths |
| q | Glottal Stop | uh-oh [ah q ow] |
| vf | Vocal Fry | Not stable, put it in random spots and see what happens... |
| cl | Plosive Modifier | Place after a plosive consonant to remove the "pop" if it's unwanted. [w ah t cl] |
Non-English Extensions
| Phoneme | Type | X-Sampa | Explaination |
|---|---|---|---|
| ee | vowel | e | Accented 'e', like in Spanish and Japanese |
| oo | vowel | o | Closed 'o', like in Spanish and Japanese |
| uu | vowel | u | Closed 'u', like in Spanish and PT-BR |
| ux | vowel | M | Japanese 'u' |
| nn | vowel* | N\ | Japanese 'ん' |
| hx | cons. | x | Spanish 'j', like in 'ojos' [oo hx oo s] |
| rr | cons. | rr | Rolled R, like in perrito [p ee rr iy tx oo] |
| rx | cons. | R | Uvular R, like in 'Março' [m aa rx s] |
| tx | cons. | t | Unaspirated t. For Non-English languages |
*: While in theory it is not a vowel, to ensure proper functionality it is treated as a vowel by DiffSinger.