-
Notifications
You must be signed in to change notification settings - Fork 5k
Closed
Description
I found an issue that's causing a core-dump with VAD, the audio file being processes is fine and I can re-process it and it works the second time, it's a stereo PCM wav file.
The whisper-server startup command is:
bin/whisper-server -m models/ggml-large-v3.bin --port 8080 -nc -sns --diarize -t 8 --vad --vad-model models/ggml-silero-v5.1.2.bin
Here's the log output prior to the core dump, I also have the actual core dump if needed:
Sep 02 14:24:58 ai0 whisper-server[389693]: operator(): processing '/tmp/01b3-b408-4c3b-c91c-f82c-cfe8-40ce-e110.wav' (356251 samples, 22.3 sec), 8 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
Sep 02 14:24:58 ai0 whisper-server[389693]: whisper_full: VAD is enabled, processing speech segments only
Sep 02 14:24:58 ai0 whisper-server[389693]: whisper_vad: VAD is enabled, processing speech segments only
Sep 02 14:24:58 ai0 whisper-server[389693]: whisper_vad_segments_from_samples: detecting speech timestamps in 356251 samples
Sep 02 14:24:58 ai0 whisper-server[389693]: whisper_vad_detect_speech: detecting speech in 356251 samples
Sep 02 14:24:58 ai0 whisper-server[389693]: whisper_vad_detect_speech: n_chunks: 696
Sep 02 14:24:58 ai0 whisper-server[389693]: whisper_vad_detect_speech: props size: 696
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_detect_speech: chunk_len: 411 < n_window: 512
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_detect_speech: vad time = 489470.28 ms processing 356251 samples
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_segments_from_probs: detecting speech timestamps using 696 probabilities
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_segments_from_probs: Merged 1 adjacent segments, now have 4 segments
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_segments_from_probs: Final speech segments after filtering: 4
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_segments_from_probs: VAD segment 0: start = 0.64, end = 1.05 (duration: 0.41)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_segments_from_probs: VAD segment 1: start = 1.44, end = 7.77 (duration: 6.33)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_segments_from_probs: VAD segment 2: start = 8.00, end = 21.60 (duration: 13.60)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad_segments_from_probs: VAD segment 3: start = 21.92, end = 22.27 (duration: 0.35)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: detected 4 speech segments
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: Including segment 0: 0.64 - 1.15 (duration: 0.51)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: Including segment 1: 1.44 - 7.87 (duration: 6.43)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: Including segment 2: 8.00 - 21.70 (duration: 13.70)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: Including segment 3: 21.92 - 22.27 (duration: 0.35)
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: total duration of speech segments: 20.99 seconds
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: vad_segment_info: orig_start: 0.64, orig_end: 1.05, vad_start: 0.00, vad_end: 0.51
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: vad_segment_info: orig_start: 1.44, orig_end: 7.77, vad_start: 0.61, vad_end: 7.04
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: vad_segment_info: orig_start: 8.00, orig_end: 21.60, vad_start: 7.14, vad_end: 20.84
Sep 02 14:24:59 ai0 audit[389693]: ANOM_ABEND auid=4294967295 uid=0 gid=0 ses=4294967295 subj=system_u:system_r:unconfined_service_t:s0 pid=389693 comm="whisper-server" exe="/opt/whisper/build/bin/whisper-server" sig=6 res=1
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: vad_segment_info: orig_start: 21.92, orig_end: 22.27, vad_start: 20.94, vad_end: 21.29
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: Created time mapping table with 106 points
Sep 02 14:24:59 ai0 whisper-server[389693]: whisper_vad: Reduced audio from 356251 to 340571 samples (4.4% reduction)
Sep 02 14:24:59 ai0 whisper-server[389693]: malloc(): invalid size (unsorted)
theY4Kman and DeRep311
Metadata
Metadata
Assignees
Labels
No labels