Repository navigation
LTM Audio: language selection or auto-detect for transcription (non-English speech) #1095
geertvanzoest
started this conversation in
Ideas
Replies: 1 comment
|
Hey @geertvanzoest Thank you so much for reaching out to support! I will ensure to pass this feedback + feature request off to the Engineering team! Thank you again for reaching out. Please reach out if you have any other good feature requests or run into any unexpected behavior! |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Summary
LTM Audio currently transcribes all captured speech as if it were English. For users who speak another language (in my case Dutch), the resulting transcripts are unusable: the model force-fits non-English speech into English words, producing garbled text. Downstream, most of these audio events end up classified as "Unclassified Event — content could not be meaningfully analyzed", so the spoken content contributes little to workstream summaries beyond the fact that a conversation took place.
Request
Add language support to the LTM Audio transcription pipeline, ideally one (or both) of:
Context
Environment
Happy to provide more detail or test builds.
All reactions