Repository navigation
Inserting 'thank you' at the end of dictation sometimes #108
Description
HEX version
2.1.22
Platform
macOS
System details
27.2 beta
What happened?
I'm noticing that it inserts 'thank you' at the end of some sentences sometimes. I thought that maybe it heard it muttered externally but I'm in a room alone and it still does it
Steps to reproduce
- dictate
- wait for random 'thank you'
Additional context
No response
Activity
kitlangton commentedon Oct 2, 2026
Thanks for the report. Which transcription model and language do you have selected (Settings, or the model picker in the menu bar)?
A stray "thank you" at the end is a known failure mode of some speech models, especially Whisper. When a recording ends with a bit of silence or room noise, the model can "hear" a common closing phrase that was never said, because phrases like that show up at the end of a lot of its training audio. So it's very likely the model and not a sound in your room.
If you're on Whisper large-v3-turbo, could you try Parakeet (Unified English or v3 if you dictate in English) and let us know whether it still happens? It also helps to know whether it shows up more after you pause before releasing the shortcut, and whether OpenCode rewriting is enabled in the Mode you're using.
koalazub commentedon Oct 4, 2026
Excuse the delay in getting back to you. So this was for the Cohere transcribe(English) model:
I notice that this happens if I were to press and hold the transcribe key and then let go without adding any input. It would insert thank you automatically. And then on occasion, sometimes when I'm dictating.
feat: Hex 2.2.0 — faster dictation and the Meetings beta (#133)
kitlangton commentedon Oct 18, 2026
2.2.0 is out and now discards near-silent recordings before they reach the model, which should stop the "thank you" you got from holding and releasing without speaking.
The occasional "thank you" during real speech isn't specifically addressed yet. I'm closing this for now, but if it still happens on 2.2.0, please reopen and mention the model and roughly how long the dictation was.