PR #20722 opened by Zhao Zhili (quink) URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20722 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20722.patch
>From bb65f51fce43d970e94eeef12dd20d3beabb559e Mon Sep 17 00:00:00 2001 From: Zhao Zhili <[email protected]> Date: Fri, 15 Aug 2025 20:39:49 +0800 Subject: [PATCH 1/5] avfilter/af_whisper: fix broken output for multibyte character text + 1 can break a multibyte character, e.g., Chinese in UTF-8. There is no space at the beginning in this case. (cherry picked from commit 1d06e8ddcd8c14232d0a3c2b1c21e50b232549b4) --- libavfilter/af_whisper.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/libavfilter/af_whisper.c b/libavfilter/af_whisper.c index 186b624504..bd145d6d1d 100644 --- a/libavfilter/af_whisper.c +++ b/libavfilter/af_whisper.c @@ -215,7 +215,9 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples) for (int i = 0; i < n_segments; ++i) { const char *text = whisper_full_get_segment_text(wctx->ctx_wsp, i); - char *text_cleaned = av_strireplace(text + 1, "[BLANK_AUDIO]", ""); + if (av_isspace(text[0])) + text++; + char *text_cleaned = av_strireplace(text, "[BLANK_AUDIO]", ""); if (av_strnlen(text_cleaned, 1) == 0) { av_freep(&text_cleaned); -- 2.49.1 >From b784c3eb6d4f989b9e78f2d93e1b4d73018a1f4f Mon Sep 17 00:00:00 2001 From: Gyan Doshi <[email protected]> Date: Sat, 16 Aug 2025 16:12:10 +0530 Subject: [PATCH 2/5] avfilter/whisper: correct option formatting (cherry picked from commit 7df92712723dee0ace3596684a90282c0e12a8ef) --- libavfilter/af_whisper.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/libavfilter/af_whisper.c b/libavfilter/af_whisper.c index bd145d6d1d..385180b4ed 100644 --- a/libavfilter/af_whisper.c +++ b/libavfilter/af_whisper.c @@ -430,7 +430,8 @@ static int query_formats(const AVFilterContext *ctx, #define HOURS 3600000000 static const AVOption whisper_options[] = { - { "model", "Path to the whisper.cpp model file", OFFSET(model_path), AV_OPT_TYPE_STRING,.flags = FLAGS }, { "language", "Language for transcription ('auto' for auto-detect)", OFFSET(language), AV_OPT_TYPE_STRING, {.str = "auto"}, .flags = FLAGS }, + { "model", "Path to the whisper.cpp model file", OFFSET(model_path), AV_OPT_TYPE_STRING,.flags = FLAGS }, + { "language", "Language for transcription ('auto' for auto-detect)", OFFSET(language), AV_OPT_TYPE_STRING, {.str = "auto"}, .flags = FLAGS }, { "queue", "Audio queue size", OFFSET(queue), AV_OPT_TYPE_DURATION, {.i64 = 3000000}, 20000, HOURS, .flags = FLAGS }, { "use_gpu", "Use GPU for processing", OFFSET(use_gpu), AV_OPT_TYPE_BOOL, {.i64 = 1}, 0, 1, .flags = FLAGS }, { "gpu_device", "GPU device to use", OFFSET(gpu_device), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX, .flags = FLAGS }, -- 2.49.1 >From adc819773ba98db56bbff0e4da8187e6ed3d7db9 Mon Sep 17 00:00:00 2001 From: Vittorio Palmisano <[email protected]> Date: Fri, 29 Aug 2025 11:32:20 +0200 Subject: [PATCH 3/5] avfilter/af_whisper: fix srt file format The SRT file format requires commas in the time string, not periods. (cherry picked from commit 73d411c399df4abe2750b611fc8381979fcbafc6) --- libavfilter/af_whisper.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavfilter/af_whisper.c b/libavfilter/af_whisper.c index 385180b4ed..663fe446bb 100644 --- a/libavfilter/af_whisper.c +++ b/libavfilter/af_whisper.c @@ -246,7 +246,7 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples) if (!av_strcasecmp(wctx->format, "srt")) { buf = av_asprintf - ("%d\n%02ld:%02ld:%02ld.%03ld --> %02ld:%02ld:%02ld.%03ld\n%s\n\n", + ("%d\n%02ld:%02ld:%02ld,%03ld --> %02ld:%02ld:%02ld,%03ld\n%s\n\n", wctx->index, start_t / 3600000, (start_t / 60000) % 60, (start_t / 1000) % 60, start_t % 1000, end_t / 3600000, (end_t / 60000) % 60, -- 2.49.1 >From d8049e01d79df68520371fc8e0a93c5dd786df06 Mon Sep 17 00:00:00 2001 From: Vittorio Palmisano <[email protected]> Date: Sun, 21 Sep 2025 15:19:51 +0200 Subject: [PATCH 4/5] avfilter/af_whisper: fix int64 printf format Use PRId64 for printing int64_t values in the SRT output. (cherry picked from commit f18b1e23890c1179d41c07f2c195c56fb90ea072) --- libavfilter/af_whisper.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/libavfilter/af_whisper.c b/libavfilter/af_whisper.c index 663fe446bb..90b2fc89a0 100644 --- a/libavfilter/af_whisper.c +++ b/libavfilter/af_whisper.c @@ -150,7 +150,7 @@ static int init(AVFilterContext *ctx) } av_log(ctx, AV_LOG_INFO, - "Whisper filter initialized: model: %s lang: %s queue: %ld ms\n", + "Whisper filter initialized: model: %s lang: %s queue: %" PRId64 " ms\n", wctx->model_path, wctx->language, wctx->queue / 1000); return 0; @@ -194,7 +194,7 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples) const float duration = (float) samples / WHISPER_SAMPLE_RATE; av_log(ctx, AV_LOG_INFO, - "run transcription at %ld ms, %d/%d samples (%.2f seconds)...\n", + "run transcription at %" PRId64 " ms, %d/%d samples (%.2f seconds)...\n", timestamp_ms, samples, wctx->audio_buffer_fill_size, duration); struct whisper_full_params params = whisper_full_default_params(WHISPER_SAMPLING_GREEDY); @@ -228,7 +228,7 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples) 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; - av_log(ctx, AV_LOG_DEBUG, " [%ld-%ld%s]: \"%s\"\n", + av_log(ctx, AV_LOG_DEBUG, " [%" PRId64 "-%" PRId64 "%s]: \"%s\"\n", timestamp_ms + t0_ms, timestamp_ms + t1_ms, turn ? " (turn)" : "", text_cleaned); if (segments_text) { @@ -246,13 +246,13 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples) if (!av_strcasecmp(wctx->format, "srt")) { buf = av_asprintf - ("%d\n%02ld:%02ld:%02ld,%03ld --> %02ld:%02ld:%02ld,%03ld\n%s\n\n", + ("%d\n%02" PRId64 ":%02" PRId64 ":%02" PRId64 ",%03" PRId64 " --> %02" PRId64 ":%02" PRId64 ":%02" PRId64 ",%03" PRId64 "\n%s\n\n", wctx->index, start_t / 3600000, (start_t / 60000) % 60, (start_t / 1000) % 60, start_t % 1000, end_t / 3600000, (end_t / 60000) % 60, (end_t / 1000) % 60, end_t % 1000, text_cleaned); } else if (!av_strcasecmp(wctx->format, "json")) { - buf = av_asprintf("{\"start\":%ld,\"end\":%ld,\"text\":\"%s\"}\n", start_t, end_t, text_cleaned); + buf = av_asprintf("{\"start\":%" PRId64 ",\"end\":%" PRId64 ",\"text\":\"%s\"}\n", start_t, end_t, text_cleaned); } else buf = av_strdup(text_cleaned); -- 2.49.1 >From 7cbc26267b6968eb2b8cc5f70e3603b6ee9b2101 Mon Sep 17 00:00:00 2001 From: Vittorio Palmisano <[email protected]> Date: Sun, 21 Sep 2025 15:33:17 +0200 Subject: [PATCH 5/5] avfilter/af_whisper: fix srt index The srt index should be incremented for each segment. (cherry picked from commit 9970dc32bf85628e53ed0952d87d384080d8976e) --- libavfilter/af_whisper.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/libavfilter/af_whisper.c b/libavfilter/af_whisper.c index 90b2fc89a0..3c0eba42f0 100644 --- a/libavfilter/af_whisper.c +++ b/libavfilter/af_whisper.c @@ -251,6 +251,8 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples) (start_t / 60000) % 60, (start_t / 1000) % 60, start_t % 1000, end_t / 3600000, (end_t / 60000) % 60, (end_t / 1000) % 60, end_t % 1000, text_cleaned); + + wctx->index++; } else if (!av_strcasecmp(wctx->format, "json")) { buf = av_asprintf("{\"start\":%" PRId64 ",\"end\":%" PRId64 ",\"text\":\"%s\"}\n", start_t, end_t, text_cleaned); } else @@ -265,8 +267,6 @@ static void run_transcription(AVFilterContext *ctx, AVFrame *frame, int samples) av_freep(&text_cleaned); } - wctx->index++; - AVDictionary **metadata = &frame->metadata; if (metadata && segments_text) { av_dict_set(metadata, "lavfi.whisper.text", segments_text, 0); -- 2.49.1 _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
