Alpha Cephei
The speech company behind vosk, founded and run by nickolay-shmyrev. Small, open by default,
and unusual in this spoke for shipping across both branches: Vosk for recognition and Vosk-TTS
for synthesis, with the evaluation harness for the latter published alongside it
(alphacep/vosk-tts).
vosk holds the small/embedded, permissively-licensed pole of the open STT field here — ~50 MB models, 20+ languages, Apache-2.0, Raspberry Pi to cluster. The same instinct shows in the TTS work: Vosk-TTS is multi-speaker and CPU-first (xRT 0.0605 on CPU in russian-tts-evaluation) rather than a scale play.
Publishing the benchmark and competing in it is the thing worth recording. Most of the quality claims in this spoke arrive from vendors with no method attached — luxtts asserting parity with models ten times larger, supertonic-2's vendor table, audio8-tts's unmeasured INT8. Alpha Cephei published the metrics, the test-set description and the scoring code, then entered its own system into the comparison. That is a better position than the field's norm and still a conflict of interest; russian-tts-evaluation treats it as both.
Connections
- founded and run by nickolay-shmyrev
- publisher of vosk (STT) and Vosk-TTS (synthesis)
- publisher of russian-tts-evaluation
Related
vosk · nickolay-shmyrev · russian-tts-evaluation · russian-tts