Microsoft · 2020 · variance predictors

FastSpeech 2

FastSpeech with explicit pitch, energy, and duration predictors, trained on real targets rather than a teacher model. Still the skeleton inside a lot of production non-AR stacks.

Known for

Controllable parallel TTS without a Tacotron teacher.

Vendor / paper demo

LJSpeech · FastSpeech 2 official demo, sample 0000000004.

Were the leaders in this luckless change, though our own Baskerville, who was at work some years before them, went much on the same lines.

Year / era
2020
License class
research
Cloning
No
Streaming
No
Emotion controls
Yes
Languages (tags)
en
Official
Source
Kokoro 82M · af_heart