Vendor / paper demo
LJSpeech · FastSpeech official demo, sample 1. Same sentence as Merlin and Tacotron 2 on that page.
I will quote an extract from the reverend gentleman's own journal.
Microsoft · 2019 · parallel AR-free
FastSpeech
Throw away autoregressive decoding. Predict duration, then generate mel frames in parallel. Faster, more stable, slightly less magic. The clip is sample 1 from the paper demo, next to Merlin on the same page.
Known for
Making neural TTS fast enough to feel like a vocoder problem again.
- Approach
- Parallel / non-AR
- Year / era
- 2019
- License class
- research
- Cloning
- No
- Streaming
- No
- Emotion controls
- No
- Languages (tags)
- en
- Official
- Source