Show HN: WhatYouSay – Self-hosted WhatsApp voice-note transcription (github.com)

WhatYouSay, a self-hosted Node.js server application, has been announced, which enables real-time transcription of WhatsApp voice notes directly into text replies within chats. By linking to a WhatsApp account as an unofficial device, the app transcribes voice messages using the Whispers transcription model and cleans up the text with a language model. Users can choose whether to transcribe sent or received notes, translate messages, and receive live updates via a web dashboard, all without storing transcripts, enhancing privacy. This project is particularly significant for the AI/ML community as it offers an independent and privacy-centric alternative to existing services like WhatScribe. The application utilizes a two-pass transcription strategy to reduce truncation errors and offers language-based cleanup mechanisms. However, it is important to note that using unofficial WhatsApp clients could lead to account restrictions. With its various technical features, including real-time processing and optional Telegram integration for alerts, WhatYouSay showcases the potential for self-hosted solutions in translating and summarizing communications in a privacy-respecting manner.