Nickolay Shmyrev
Speech-recognition engineer behind alpha-cephei and its two open projects, vosk (offline STT,
Apache-2.0, ~15k★) and Vosk-TTS. He writes the alphacephei.com/nsh/ blog, where
russian-tts-evaluation was published on 2024-07-12.
That evaluation is the reason he has a page. It is the only independent side-by-side measurement of Russian TTS this spoke has located — thirteen systems, five metrics, published scoring code, an audiobook test set — and everything russian-tts can say with numbers traces to it.
He is also a party to it: Vosk-TTS is his own system and takes the best intonation and studio-quality scores in his own table. The page records that beside the results rather than behind them.
His stated methodological position, from the same post, is one this spoke keeps meeting from other directions: clarity outranks intonation for a listener, because "you can deal with plain intonation but artifacts are really annoying." He is candid about the instrument's limits too, writing that FAD "only works for multivoice systems" — an admission that the intonation column is weakest for precisely the single-speaker systems that win the clarity column.
Connections
- author of russian-tts-evaluation
- founder of alpha-cephei; maintainer of vosk and Vosk-TTS