Сергей Дуканов, Тинькофф — Text to speech: cooking recipe

In this talk, Sergey examined the speech synthesis service pipeline and discussed each of its components: how to implement it, the challenges that arise, and how to overcome them. He also tried to demonstrate that speech synthesis presents a variety of interesting challenges at the intersection of SotA DL architectures and engineering. In this video: — What is TTS and why is it needed? — What is an acoustic model? — The main challenges of teaching an acoustic model? Digests, articles, and meetup announcements: https://t.me/itstinkoff Blog on Habr: https://l.tinkoff.ru/habr-tinkoff Our vacancies: https://l.tinkoff.ru/its-career Subscribe to the IT's Tinkoff channel:    / itstinkoff   00:00 — Speech synthesis in Tinkoff 02:01 — Problems on the path from text to sound 03:26 — Problem solving: text normalization 09:41 — Intonation tags 13:35 — Acoustic model 16:29 — Challenges of teaching an acoustic model 18:40 — Experiments with an acoustic model 25:27 — Acoustic Model Quality Metrics 28:11 — Summary and Q&A #tts #stt #asr #dl #deeplearning #voicetech #voice #text-to-speech #speech_synthesis