TTS-Framework docs
This is the documentation for the TTS-Framework. Here you can find the technical description of the project and the API reference. Also, here you can find the code specs, the project structure and the development guidelines.
Basically here you can find everything you need to know about the project.
You can read about this mkdocs-material here: mkdocs-material
[WIP] Modified version of DelightfulTTS and UnivNet codec.
References
Development docs
Description of the training process. Docs, ideas and examples for the training process.
Models
TTS Models
DelightfulTTS
The DelightfulTTS: The Microsoft Speech Synthesis System for Blizzard Challenge 2021
Vocoder
Univnet
Most neural vocoders employ band-limited mel-spectrograms to generate waveforms. If full-band spectral features are used as the input, the vocoder can be provided with as much acoustic information as possible. However, in some models employing full-band mel-spectrograms, an over-smoothing problem occurs as part of which non-sharp spectrograms are generated. To address this problem, we propose UnivNet, a neural vocoder that synthesizes high-fidelity waveforms in real time. Inspired by works in the field of voice activity detection, we added a multi-resolution spectrogram discriminator that employs multiple linear spectrogram magnitudes computed using various parameter sets. Using full-band mel-spectrograms as input, we expect to generate high-resolution signals by adding a discriminator that employs spectrograms of multiple resolutions as the input. In an evaluation on a dataset containing information on hundreds of speakers, UnivNet obtained the best objective and subjective results among competing models for both seen and unseen speakers. These results, including the best subjective score for text-to-speech, demonstrate the potential for fast adaptation to new speakers without a need for training from scratch.
Training
Training code documentation.