whisper.cpp
vad : add initial Voice Activity Detection (VAD) support
#3065
Merged

vad : add initial Voice Activity Detection (VAD) support #3065

danbev merged 49 commits into ggml-org:master from danbev:vad
danbev
danbev danbev force pushed from 0f2fe06f to b10e6ddd 1 year ago
danbev danbev force pushed from 4aea8b3a to 3cca1a23 1 year ago
danbev danbev force pushed from 5758650f to 9f0ed3d8 1 year ago
tannisroot
danbev
danbev danbev force pushed from 63d3fe24 to ebc79f8a 1 year ago
danbev danbev force pushed from b59768bb to 798695fc 1 year ago
danbev danbev marked this pull request as ready for review 1 year ago
ggerganov
danbev
danbev danbev force pushed from 798695fc to 6b56b7df 1 year ago
danbev
mrfragger
ggerganov
ggerganov commented on 2025-05-07
ggerganov
TeslaKang
ggerganov ggerganov force pushed from 5a6236e9 to 60d561b8 1 year ago
ggerganov
danbev
ggerganov
ggerganov commented on 2025-05-09
danbev
danbev commented on 2025-05-09
danbev vad : add initial Voice Activity Detection (VAD) support
871da0bf
danbev examples : add VAD parameters to CLI [no ci]
24901683
danbev ci : add job to test VAD
eb23253b
danbev vad : map timestamps to original audio
59252c2b
danbev squash! vad : add initial Voice Activity Detection (VAD) support [no ci]
37a36a33
danbev vad : extract VAD processing to a separate function
033c0ce2
danbev vad : add TODOs to optimize segment access [no ci]
028481e9
danbev vad : only use CPU backend for VAD processing [no ci]
fc7ebf20
danbev tests : fix strcmp assert and use beam search
3276232e
danbev vad : dont reshape stft_forward_basis tensor
abc05c5c
danbev vad : use ggml_row_size() and rename hdim_bytes to hdim_size
0e18ceba
danbev vad : remove unnecessary ggml_cont
9bf1b4b3
danbev vad : fix typo in log message
dc529950
danbev vad : don't use left leaning ref for segment
2b057733
danbev vad : use std::vector<float> instead float pointers
44bdef1b
danbev vad : enable GPU support for VAD but default to false
27eb59bd
danbev vad : use kebab-case and not snake_case for VAD options
643a91bf
danbev vad : add h_state and c_state to whisper_vad_state
e4d43072
danbev vad : always initialize filtered_n_samples to 0
94c3aba8
danbev vad : use orig timestamp for first segment
e70e4861
ggerganov vad : fix buffers and enable GPU support by default
436baeb7
danbev vad : fix use_gpu assert in test-vad.cpp
eb2c83ee
danbev vad : remove unnecessary reserve [no ci]
47c8f02d
danbev vad : add probs to whisper_vad_state
327cdaee
danbev vad : add timing of vad processing [no ci]
bf2b0df9
danbev danbev force pushed from 50337e26 to bf2b0df9 1 year ago
danbev
enesgrahovac
ggerganov vad : force GPU off for now
243e0dba
ggerganov vad : minor style and naming changes
65c421de
ggerganov vad : minor style
cae38fda
ggerganov vad : remove obsolete whisper_vad_free_speech
cd953ebf
ggerganov vad : refactor whiser_vad_params API
f42e6e47
ggerganov vad : simplify whisper_vad_timestamps_from_probs()
4ff858ba
ggerganov vad : refactor whisper_vad_timestamps_from_probs to use C++
13a75177
ggerganov ggerganov force pushed from 66bc585e to 13a75177 1 year ago
ggerganov
danbev vad : make whisper_vad_timestamps oblique in API
3bcc44c2
danbev vad : rename whisper_vad_speech to whisper_vad_probs
5543c80c
ggerganov
ggerganov commented on 2025-05-10
danbev vad : move whisper_vad_segment to whisper.cpp
8b6f19cf
danbev vad : make segments vector a std::vector
7625ba16
danbev vad : use std::vector for segments in whisper_vad_timestamps_from_probs
b0b2f9b4
danbev vad : rename pcmf32 parameters to samples [no ci]
20fe0b35
ggerganov
ggerganov commented on 2025-05-10
ggerganov
ggerganov commented on 2025-05-10
ggerganov
ggerganov commented on 2025-05-10
danbev vad : remove n_segments from struct whisper_vad_timestamps
f2123105
MahmoudAshraf97
danbev vad : rename whisper_vad_timestamps to whisper_vad_segments [no ci]
4c7fe00c
danbev
danbev vad : remove whisper_vad_probs struct [no ci]
dc541f93
danbev vad : remove whisper_vad_state struct
163ad538
TeslaKang
danbev vad : remove window_size_samples from VAD params
810981f4
danbev vad : clarify VAD CLI options [no ci]
050038ca
danbev docs : add VAD section to README.md [no ci]
3cff6587
danbev squash! docs : add VAD section to README.md [no ci]
acc8747d
ggerganov vad : minor rename
7aac6eca
ggerganov
ggerganov approved these changes on 2025-05-12
ggerganov
ggerganov commented on 2025-05-12
danbev squash! docs : add VAD section to README.md [no ci]
41c20100
danbev vad : fix cli option names [no ci]
67f0fd40
danbev danbev merged e41bc5c6 into master 1 year ago
vrs
danbev
mdestagnol
danbev
ggerganov
mdestagnol
danbev
WilliamTambellini
stephantucker992-prog
stephantucker992-prog commented on 2025-10-01

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone