SKTTS research checkpoints
Research checkpoints of the SKTTS project (an open TTS; English first, Slovak later). Work in progress: these are intermediate training states, not a finished product.
| Folder | What | Training data |
|---|---|---|
ae64x/ |
44.1 kHz speech autoencoder, 64-dim latent: encoder from a mel-only run, upsampling decoder taken from ae44d and GAN-trained 14k more steps (best so far) |
HiFi-TTS, VCTK |
aelat/<size>/ |
latent-size test (24/48/64), short mel-only runs | HiFi-TTS, VCTK |
ae44a/, ae44d/ |
earlier 44.1 kHz autoencoders, 24-dim latent | HiFi-TTS, VCTK |
v2-en/, v2-en-long/, v21-en/ |
English acoustic models (Matcha-style, v2 architecture) | LibriTTS-R, HiFi-TTS |
matcha-en-baseline/ |
English Matcha baseline | LibriTTS-R |
analysis/ |
per-clip quality analysis of the Slovak source datasets | TrePal, lili |
Data and licences
- HiFi-TTS (
MikhailT/hifi-tts), CC BY 4.0 - VCTK (
sanchit-gandhi/vctk), CC BY 4.0 - LibriTTS-R (
mythicinfinity/libritts_r), CC BY 4.0 - TrePal (
streams-sro/trepal, recordings of the National Council of the Slovak Republic), CC BY 4.0 - lili (
vsisik/voice-dataset-lili, from NabuCasa voice-datasets), CC0
The checkpoints are released under CC BY 4.0 with attribution to the datasets above. Do not use them to imitate the voice of a real person.