[MM-54242] Improve timestamp accuracy #3
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
streamer45
added
2: Dev Review
cpoile
approved these changes
Nov 10, 2023
Comment on lines
+352
to
+353
| startSampleOff := int(seg.SpeechStartAt * trackOutAudioRate) | ||
| endSampleOff := int(seg.SpeechEndAt * trackOutAudioRate) |
There was a problem hiding this comment.
One request -- can you add some comments for future us about what the units are for seg.SpeechStartAt and startSampleOff? (That will help in the future when I'm reviewing and remembering what the numbers mean.)
There was a problem hiding this comment.
Sure but the sample offset is just what it says. It's an index to the first sample where the speech starts (+- some padding), there's no unit really.
jupenur
requested changes
Nov 14, 2023
* Implement more human friendly filenames * Include Silero VAD model v4 * Cache CGO dependencies and Whisper models in Docker build * Update silero-vad-go * Add SHA check for ONNXRuntime * Support language autodetection * Initial multi-threading support * Fix marshaling case * Tune speech detector silence duration threshold * Sanitize Text and Speaker strings * Build as position-independent executable * Update rtcd client dependency * Better escaping
jupenur
approved these changes
Nov 15, 2023
streamer45
added
3: Reviews Complete
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
PR adds a speech detection step prior to transcribing. This was made necessary as
whisper.cppdoesn't currently support word level timestamps (see ggerganov/whisper.cpp#375) so portions of silence could skew the sync of the final transcription by quite a bit.The way this works is by first feeding the decoded audio samples to the Silero VAD model and essentially trim silence from the track, splitting it into separate speech segments before passing it to the Whisper engine for transcription.
@jupenur Apologies for the late addition of a non trivial dependency. Let me know if you have any questions.
Ticket Link
https://mattermost.atlassian.net/browse/MM-54242