Skip to content

The truly free, open-source, local-first, cross-platform CapCut replacement. Scriptable through a JSON-RPC API & MCP.

fix: stamp mix input PTS from sample count, ignore source timestamps

Domica6d1274145b16
diff --git a/src/crates/concat-media/src/audio.rs b/src/crates/concat-media/src/audio.rs
index c4cb0fd7..60b78a2f 100644
--- a/src/crates/concat-media/src/audio.rs
+++ b/src/crates/concat-media/src/audio.rs
@@ -225,10 +225,11 @@ pub fn mix_graph(clips: &[AudioClip], duration: f64) -> Result<String> {
// based trim close the input early. `mix_to_file` counts the source
// samples it feeds per input and stops at `duration * speed` source
// seconds instead; see the `clip_samples` field.
- let mut stage = format!(
- "[{index}:a]atrim=start={:.6},asetpts=PTS-STARTPTS",
- clip.source_start
- );
+ // No trim or restamp in the graph: Rust stamps each frame's PTS from
+ // a sample counter and stops feeding the input after duration*speed
+ // source seconds. This sidesteps timestamp discontinuities (edit
+ // lists, paused recordings) that parked frames outside the mix.
+ let mut stage = format!("[{index}:a]anull");
for filter in speed_filters(speed, clip.preserve_pitch) {
stage.push(',');
@@ -393,7 +394,6 @@ pub fn mix_to_file(clips: &[AudioClip], duration: f64, destination: &Path) -> Re
}
})?;
let stream = input.stream(stream_index).expect("just found");
- let time_base = stream.time_base();
let decoder = ffmpeg::codec::Context::from_parameters(stream.parameters())
.and_then(|context| context.decoder().audio())
.map_err(|error| ffi::fail("open decoder", path, error))?;
@@ -413,9 +413,11 @@ pub fn mix_to_file(clips: &[AudioClip], duration: f64, destination: &Path) -> Re
samples_sent: 0,
clip_samples: 0,
};
- let first = mix_input.next()?.ok_or_else(|| Error::NoAudioStream {
+ let mut first = mix_input.next()?.ok_or_else(|| Error::NoAudioStream {
path: path.to_path_buf(),
})?;
+ // Our continuous PTS: sample count from the start of this clip.
+ first.set_pts(Some(0));
// The out-point lives here, not in the filtergraph: how many source
// samples this clip may feed. A sped-up clip covers more source than
// its timeline length, hence the speed factor. The first frame goes
@@ -423,10 +425,11 @@ pub fn mix_to_file(clips: &[AudioClip], duration: f64, destination: &Path) -> Re
mix_input.samples_sent = first.samples() as i64;
mix_input.clip_samples =
(clip.duration * clamp_speed(clip.speed) * first.rate() as f64).round() as i64;
+ // One sample per time-base tick: the PTS we stamp is a plain
+ // running sample count, so the graph's clock is the sample clock.
let args = format!(
- "time_base={}/{}:sample_rate={}:sample_fmt={}:channel_layout={}",
- time_base.numerator(),
- time_base.denominator(),
+ "time_base=1/{}:sample_rate={}:sample_fmt={}:channel_layout={}",
+ first.rate(),
first.rate(),
first.format().name(),
first.ch_layout().description()
@@ -595,8 +598,12 @@ pub fn mix_to_file(clips: &[AudioClip], duration: f64, destination: &Path) -> Re
continue;
}
match input.next()? {
- Some(frame) => {
+ Some(mut frame) => {
let frame_samples = frame.samples() as i64;
+ // Our continuous PTS: whatever the source's timestamp
+ // series does from here on cannot move this frame out
+ // of its place in the mix.
+ frame.set_pts(Some(input.samples_sent));
// Stop feeding this input once we've sent its clip's worth.
// The graph's apad will fill the rest with silence until
// the mix's atrim=duration ends the whole thing.