We had a breakthrough! Synapse can now horizontally scale across multiple Python processes, with message sending no longer going via the main proc. This has spectacularly improved perf on http://Matrix.org over the last few weeks🎉 Read all about it: https://matrix.org/blog/2020/11/03/how-we-fixed-synapses-scalability
Speech recognition kit Vosk has supported quick reconfiguration of the vocabulary at runtime, so the pronunciation or audio quality doesn't need to be perfect as the program will try to match the word with the most similar sound in the given list. This makes Vosk suitable for building anything that accepts voice commands.
Rust binding https://github.com/wzhd/vosk-rs just updated this feature to support a phrase list instead of a word list . (Based on upstream commit: https://github.com/alphacep/vosk-api/commit/564fab7ec1ea097916a4251530ad151fef940c1f)
Voice recognition with Rust running locally on a core of a 6-year old Intel CPU generating subtitles in real-time (https://github.com/wzhd/vosk-rs)