Hello everyone,
This is my first project for Home Assistant, and I’m not a professional developer, so I’m quite nervous about the quality (or lack thereof) of the source code. Nevertheless, it works.
This project attempts to combine or leverage the phrase correction from Wyoming Vosk with more powerful Wyoming-compatible speech-to-text services like Faster Whisper, Google Cloud, or Azure Cloud to achieve better results. It’s like a middle-ground—and very basic—version between Speech to Phrase and a fully open speech-to-text service.
I’m currently using it in combination with GitHub - hugobloem/wyoming-microsoft-stt: Wyoming protocol server for Microsoft Azure speech-to-text, which has given me very good results. Microsoft’s STT is really good, but it occasionally fails when there’s background noise. This is where sentence correction has helped me a lot in getting a speech-to-text output that’s suitable for daily use.
Repository and instructions here: GitHub - Cheerpipe/wyoming_rapidfuzz_proxy
Warnings:
- The project works, but the code quality may be quite low. All kinds of suggestions for improvement are gladly welcome.
- You need to know what you’re doing, or you might overly restrict your voice commands and ruin the assistant.
- If the number of statements and variations grows too large, the correction process database can become excessively big—even hundreds of gigabytes.