github-mirrors/FFmpeg
1
0
Fork 0
mirror of https://mirror.skon.top/https://github.com/FFmpeg/FFmpeg synced 2026-04-20 21:00:41 +08:00
Code Issues Packages Projects Releases Wiki Activity

avfilter/af_whisper: Add max_len parameter

Browse Source
This commit is contained in:
WyattBlue
2025-12-21 23:51:15 -05:00
committed by Marton Balint
parent 75f5d79f6a
commit e48eaa8c62
2 changed files with 18 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
doc/filters.texi
View File

@@ -7768,6 +7768,12 @@ The destination format string; it could be "text" (only the transcribed text
will be sent to the destination), "srt" (subtitle format) or "json".
Default value: @code{"text"}
@item max_len
Maximum segment length in characters. When set to a value greater than 0,
transcription segments will be split to not exceed this length. This is useful
for generating subtitles with shorter lines.
Default value: @code{"0"}
@item vad_model
Path to the VAD model file. If set, the filter will load an additional voice
activity detection module (https://github.com/snakers4/silero-vad) that will be

12
libavfilter/af_whisper.c
View File

@@ -52,6 +52,7 @@ typedef struct WhisperContext {
int64_t queue;
char *destination;
char *format;
int max_len;
struct whisper_context *ctx_wsp;
struct whisper_vad_context *ctx_vad;
@@ -204,6 +205,8 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples)
params.print_progress = 0;
params.print_realtime = 0;
params.print_timestamps = 0;
params.max_len = wctx->max_len;
params.token_timestamps = (wctx->max_len > 0);
if (whisper_full(wctx->ctx_wsp, params, wctx->audio_buffer, samples) != 0) {
av_log(ctx, AV_LOG_ERROR, "Failed to process audio with whisper.cpp\n");
@@ -224,6 +227,14 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples)
continue;
}
// Skip segments that are parts of [BLANK_AUDIO] when max_len splits them
if (wctx->max_len > 0 && (strcmp(text_cleaned, "[") == 0 || strcmp(text_cleaned, "]") == 0 ||
strcmp(text_cleaned, "BLANK") == 0 || strcmp(text_cleaned, "_") == 0 ||
strcmp(text_cleaned, "AUDIO") == 0)) {
av_freep(&text_cleaned);
continue;
}
const bool turn = whisper_full_get_segment_speaker_turn_next(wctx->ctx_wsp, i);
const int64_t t0_ms = whisper_full_get_segment_t0(wctx->ctx_wsp, i) * 10;
const int64_t t1_ms = whisper_full_get_segment_t1(wctx->ctx_wsp, i) * 10;
@@ -437,6 +448,7 @@ static const AVOption whisper_options[] = {
{ "gpu_device", "GPU device to use", OFFSET(gpu_device), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX, .flags = FLAGS },
{ "destination", "Output destination", OFFSET(destination), AV_OPT_TYPE_STRING, {.str = ""}, .flags = FLAGS },
{ "format", "Output format (text|srt|json)", OFFSET(format), AV_OPT_TYPE_STRING, {.str = "text"},.flags = FLAGS },
{ "max_len", "Max segment length in characters", OFFSET(max_len), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX, .flags = FLAGS },
{ "vad_model", "Path to the VAD model file", OFFSET(vad_model_path), AV_OPT_TYPE_STRING,.flags = FLAGS },
{ "vad_threshold", "VAD threshold", OFFSET(vad_threshold), AV_OPT_TYPE_FLOAT, {.dbl = 0.5}, 0.0, 1.0, .flags = FLAGS },
{ "vad_min_speech_duration", "Minimum speech duration for VAD", OFFSET(vad_min_speech_duration), AV_OPT_TYPE_DURATION, {.i64 = 100000}, 20000, HOURS, .flags = FLAGS },