PR #23403 opened by Jun Zhao (mypopydev) URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/23403 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/23403.patch Emit ASS {\k} centisecond overrides when burning VTT karaoke cues with the subtitles filter. Out-of-range, non-monotonic, and malformed inline timestamps are ignored; cues without valid timestamps are unchanged. ASS {\k} karaoke approximates WebVTT timed text, not ::cue(:past/:future) styling. Fixes: ticket #23289 Signed-off-by: Jun Zhao <barryjzhao@tencent.com> From 705250a38a57383f115a08e218a041d0cd878786 Mon Sep 17 00:00:00 2001 From: Jun Zhao <barryjzhao@tencent.com> Date: Mon, 8 Jun 2026 00:53:26 +0800 Subject: [PATCH] avcodec/webvttdec: map WebVTT cue timestamps to ASS {\k} tags Emit ASS {\k} centisecond overrides when burning VTT karaoke cues with the subtitles filter. Out-of-range, non-monotonic, and malformed inline timestamps are ignored; cues without valid timestamps are unchanged. ASS {\k} karaoke approximates WebVTT timed text, not ::cue(:past/:future) styling. Fixes: ticket #23289 Signed-off-by: Jun Zhao <barryjzhao@tencent.com> --- libavcodec/webvttdec.c | 77 ++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 75 insertions(+), 2 deletions(-) diff --git a/libavcodec/webvttdec.c b/libavcodec/webvttdec.c index 90675bc660..89eb9e7fdf 100644 --- a/libavcodec/webvttdec.c +++ b/libavcodec/webvttdec.c @@ -48,9 +48,61 @@ static const struct { {"u", "{\\u1}"}, {"/u", "{\\u0}"}, }; -static int webvtt_event_to_ass(AVBPrint *buf, const char *p) +/* parse a WebVTT timestamp string (HH:MM:SS.mmm or MM:SS.mmm). + * Returns milliseconds or -1 on failure. */ +static int64_t parse_webvtt_timestamp(const char *buf) +{ + int h = 0, m = 0, s = 0, ms = 0; + + if (sscanf(buf, "%d:%2d:%2d.%3d", &h, &m, &s, &ms) == 4) { + if (m > 59 || s > 59) + return -1; + return (int64_t)h * 3600000 + m * 60000 + s * 1000 + ms; + } + if (sscanf(buf, "%2d:%2d.%3d", &m, &s, &ms) == 3) { + if (m > 59 || s > 59) + return -1; + return m * 60000 + s * 1000 + ms; + } + + return -1; +} + +static void append_k_cs(AVBPrint *buf, int64_t ms) +{ + if (ms > 0) + av_bprintf(buf, "{\\k%"PRId64"}", (ms + 5) / 10); +} + +/* validate a cue timestamp tag body: must be digits/colons/periods, + * parseable, strictly within (cue_start, cue_end), and after prev_ts. + * Returns 1 and writes to *ts_out on success, 0 on failure. */ +static int read_cue_timestamp(const char *body, int len, + int64_t cue_start, int64_t cue_end, + int64_t prev_ts, int64_t *ts_out) +{ + int64_t ts; + + if (len < 1 || !av_isdigit(body[0])) + return 0; + if ((int)strspn(body, "0123456789:.") != len) + return 0; + + ts = parse_webvtt_timestamp(body); + if (ts <= cue_start || ts >= cue_end) + return 0; + if (prev_ts >= 0 && ts <= prev_ts) + return 0; + + *ts_out = ts; + return 1; +} + +static int webvtt_event_to_ass(AVBPrint *buf, const char *p, + int64_t cue_start_ms, int64_t cue_end_ms) { int i, again = 0; + int64_t prev_ts = -1, ts; while (*p) { if (*p == '<') { @@ -59,6 +111,19 @@ static int webvtt_event_to_ass(AVBPrint *buf, const char *p) if (!tag_end) break; len = tag_end - p + 1; + + /* cue timestamp: emit {\k} for elapsed ms since cue_start + * or prev_ts; following text uses this timing */ + if (len > 2 && + read_cue_timestamp(p + 1, (int)(len - 2), + cue_start_ms, cue_end_ms, prev_ts, &ts)) { + append_k_cs(buf, ts - (prev_ts >= 0 ? prev_ts : cue_start_ms)); + prev_ts = ts; + p += len; + again = 1; + continue; + } + for (i = 0; i < FF_ARRAY_ELEMS(webvtt_valid_tags); i++) { const char *from = webvtt_valid_tags[i].from; if(!strncmp(p + 1, from, strlen(from))) { @@ -91,6 +156,11 @@ static int webvtt_event_to_ass(AVBPrint *buf, const char *p) av_bprint_chars(buf, *p, 1); p++; } + + /* trailing segment: last valid ts to cue end */ + if (prev_ts >= 0 && cue_end_ms > FFMAX(prev_ts, cue_start_ms)) + append_k_cs(buf, cue_end_ms - FFMAX(prev_ts, cue_start_ms)); + return 0; } @@ -103,7 +173,10 @@ static int webvtt_decode_frame(AVCodecContext *avctx, AVSubtitle *sub, AVBPrint buf; av_bprint_init(&buf, 0, AV_BPRINT_SIZE_UNLIMITED); - if (ptr && avpkt->size > 0 && !webvtt_event_to_ass(&buf, ptr)) + if (ptr && avpkt->size > 0 && + !webvtt_event_to_ass(&buf, ptr, + avpkt->pts, + avpkt->pts + avpkt->duration)) ret = ff_ass_add_rect(sub, buf.str, s->readorder++, 0, NULL, NULL); av_bprint_finalize(&buf, NULL); if (ret < 0) -- 2.52.0