- Who it helps
- Technical users investigating accessible voice-message handling and developers testing transcript cleanup.
- Availability
- Public project requiring Node.js 22 and the documented provider access. Its WhatsApp connection uses an unofficial library; account restrictions are a practical risk. External model usage may cost money.
Hosting and data handling
The documented pipeline sends speech to a provider for transcription and can return text in the same chat. Self-hosting does not keep every processing step on your device. The creator describes retained metadata, a temporary resend cache and a pseudonymization schedule. These details matter before introducing personal recordings.
WhatYouSay voice pipeline ↗WhatYouSay privacy and transcript testing ↗
A useful failure example
A September 17 case study compares twenty voice notes before and after changes to the cleanup step. It shows how editing colloquial speech can alter a transcript. The reference transcripts came from another model, so a higher word match in that small set does not establish human-verified accuracy. The reusable practice is to keep difficult examples and check them again after changing the cleanup instructions.
What you can do
- Read the provider and retention details, then examine the creator’s failed transcript examples before considering personal audio. Open reference ↗
Where the evidence stops
No repository code was executed and no WhatsApp messages were accessed. The small creator comparison was not independently reproduced.
Sources read. Sources checked October 22, 2026 UTC. No external software workflow, model inference or reported experiment was executed.
Sources & further reading
GitHub · implementation
WhatYouSay voice pipeline ↗
Read the public voice-message pipeline, self-hosting requirements and unofficial WhatsApp integration limits. No installation or message access.
Waldo van der Lore · page checked 2026-10-22
Web · analysis
WhatYouSay privacy and transcript testing ↗
Read the twenty-note comparison, model-produced reference transcripts, provider handling and retention. The date belongs to September catch-up.
Waldo van der Lore · page checked 2026-10-22 · Published 2026-09-17
Coverage note: The creator’s repository and September case study were read. Provider handling and the twenty-note comparison were not independently tested; current cross-platform discussion was not established.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
WhatYouSay is a self-hosted WhatsApp voice-message pipeline that transcribes audio and can clean up, translate or summarize the text.