Grisha Sterling - How We Made a 2023 Speech Synthesis Model Speak Better Than a 2018 Model
Grisha Sterling. A talk on the VITS architecture. We also discuss the modifications we made to the model's training, architecture, and inference to beat the competition and teach the model to speak better.
Presentation: https://shorturl.at/GJK12
Q&A in the Salute AI Telegram channel (for ML/DS specialists): https://t.me/+F3lTC6_N3k5kNWYy
Documentation: https://shorturl.at/iyJ49
Meetup "Salute, GigaChat!": Speech Technologies and Large Language Models