/radtts-uk

🇺🇦 Ukrainian RAD-TTS++ models (decoder + models with 3 voices) and HiFiGAN model

Apache License 2.0Apache-2.0

🇺🇦 Ukrainian RADTTS/RADTTS++ models and HiFiGAN model

Join our Speech Synthesis Group in Telegram: https://t.me/speech_synthesis_uk

Donate: https://send.monobank.ua/jar/3Saxixsdua

Overview

This repository contains links to models:

  • Decoder model for RADTTS
  • Decoder model for RADTTS++
  • Pretrained model for RADTTS
  • Pretrained model for RADTTS++ with AGAP config
  • Pretrained model for RADTTS++ with BGAP config
  • Pretrained model for RADTTS++ with DAP config

Voices are from Open Source Ukrainian Text-to-Speech datasets

These models have three voices:

  • Lada
  • Tetiana
  • Mykyta

Demo

Lada

lada.mp4

Tetiana

tetiana.mp4

Mykyta

mykyta.mp4

How to run?

Clone the repository https://github.com/egorsmkv/radtts and run the following command:

python inference.py -c config.json -r models/model_dap_84000.pt -v hifigan.pt -k hifigan_config.json \
  -t test.txt -s lada --speaker_attributes lada --speaker_text lada -o results

Download

Dropbox link to all models: https://www.dropbox.com/scl/fo/0eoipxgk16o2cnw2ymba3/h?dl=0&rlkey=7trclhuzuo6xno5n06xg4z7gd

Acknowledgement

  • Dmytro Chaplynskyi @dchaplinsky: help with access to UCU's cluster
  • Decoder model for RADTTS++ and Pretrained model for RADTTS++ with DAP config have been trained on the Cluster of Excellence UCU/Lang-uk.