Recordings start with a long silent head and clip the speaker at the tail (both OP25 and Trunk Recorder) #48

New Issue

2026-11-08T00:48:36-05:00

logan commented

2026-11-08 00:48:36 -05:00

Owner listening test, 2026-11-07, node-002 (MTA Zone 2): recordings often start with a long stretch of silence and then cut the last speaker off at the end. It happens on both the OP25+PulseAudio recorder and the Trunk Recorder backend (node-26#45 trial).

Since both backends show it, suspect a stage they share before blaming either recorder: the edge trim/upload path, C2 storage or transcoding, or the frontend/Discord player. Not yet localised. First step: pull a handful of raw uploaded files from GCS, compare head/tail against the node-side timestamps (audio_start_epoch/audio_end_epoch, which C2 ignores per DEFERRED.md), and check what audio_trim did to them.

This is the recording-quality foundation: a clipped tail loses the address or disposition, and that poisons transcription and correlation. Run drb-audio-review on any fix.

Owner listening test, 2026-11-07, node-002 (MTA Zone 2): recordings often start with a long stretch of silence and then cut the last speaker off at the end. It happens on both the OP25+PulseAudio recorder and the Trunk Recorder backend (node-26#45 trial). Since both backends show it, suspect a stage they share before blaming either recorder: the edge trim/upload path, C2 storage or transcoding, or the frontend/Discord player. Not yet localised. First step: pull a handful of raw uploaded files from GCS, compare head/tail against the node-side timestamps (audio_start_epoch/audio_end_epoch, which C2 ignores per DEFERRED.md), and check what audio_trim did to them. This is the recording-quality foundation: a clipped tail loses the address or disposition, and that poisons transcription and correlation. Run drb-audio-review on any fix.

logan added the

labels 2026-11-08 00:48:36 -05:00

logan commented

2026-11-09 00:24:24 -05:00

Owner listening test, 2026-11-09, node-002. Comparing the OP25 image pinned to gr310 5b58335 (main, src-a8111740) against the previous image.

Lead: less leading silence than before. Better.
Tail: still "a little aggressive": the end of the last word gets clipped.

So the tail-clip half of this issue is still open on the current decoder.

Owner listening test, 2026-11-09, node-002. Comparing the OP25 image pinned to gr310 5b58335 (main, src-a8111740) against the previous image. - Lead: less leading silence than before. Better. - Tail: still "a little aggressive": the end of the last word gets clipped. So the tail-clip half of this issue is still open on the current decoder.

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: logan/node-26#48