Skip to content

🎥 Make videos programmatically with React

`@remotion/media`: Pad final silence in looped audio frames

Jonny Burger51f6448c8774
diff --git a/packages/media/src/audio-extraction/extract-audio.ts b/packages/media/src/audio-extraction/extract-audio.ts
index 6e0385a6ee..e377da66f3 100644
--- a/packages/media/src/audio-extraction/extract-audio.ts
+++ b/packages/media/src/audio-extraction/extract-audio.ts
@@ -114,13 +114,36 @@ const extractAudioInternal = async ({
? (mediaDurationInSeconds ?? Infinity)
: trimAfter / fps;
if (loop && loopEndInSeconds <= loopStartInSeconds) {
- throw new Error('Cannot loop audio over an empty range');
+ throw new Error(
+ `Cannot loop audio over an empty range (${loopStartInSeconds} to ${loopEndInSeconds})`,
+ );
}
const audioDataArray: UnresampledPcmS16AudioData[] = [];
let remainingDurationInSeconds = durationInSeconds;
let segmentStartInSeconds = timeInSeconds;
let pendingSilenceInSeconds = 0;
+ const appendSilence = (
+ silenceDurationInSeconds: number,
+ numberOfChannels: number,
+ sourceSampleRate: number,
+ ) => {
+ const silenceFrames = Math.ceil(
+ fixFloatingPoint(silenceDurationInSeconds * sourceSampleRate),
+ );
+ if (silenceFrames > 0) {
+ audioDataArray.push({
+ data: new Int16Array(silenceFrames * numberOfChannels),
+ numberOfChannels,
+ numberOfFrames: silenceFrames,
+ sampleRate: sourceSampleRate,
+ timestamp: timeInSeconds * 1_000_000,
+ durationInMicroSeconds:
+ (silenceFrames / sourceSampleRate) * 1_000_000,
+ });
+ }
+ };
+
while (remainingDurationInSeconds > 0.0000000001) {
const segmentDurationInSeconds = loop
? Math.min(
@@ -235,20 +258,11 @@ const extractAudioInternal = async ({
segmentAudioData[0];
// Preserve silence at the end of a loop pass before appending audio
// from the next pass.
- const silenceFrames = Math.ceil(
- fixFloatingPoint(pendingSilenceInSeconds * sourceSampleRate),
+ appendSilence(
+ pendingSilenceInSeconds,
+ numberOfChannels,
+ sourceSampleRate,
);
- if (silenceFrames > 0) {
- audioDataArray.push({
- data: new Int16Array(silenceFrames * numberOfChannels),
- numberOfChannels,
- numberOfFrames: silenceFrames,
- sampleRate: sourceSampleRate,
- timestamp: timeInSeconds * 1_000_000,
- durationInMicroSeconds:
- (silenceFrames / sourceSampleRate) * 1_000_000,
- });
- }
audioDataArray.push(...segmentAudioData);
const segmentFrames = segmentAudioData.reduce(
@@ -271,6 +285,16 @@ const extractAudioInternal = async ({
return {data: null, durationInSeconds: mediaDurationInSeconds};
}
+ if (loop && pendingSilenceInSeconds > 0) {
+ const {numberOfChannels, sampleRate: sourceSampleRate} =
+ audioDataArray[0];
+ appendSilence(
+ pendingSilenceInSeconds,
+ numberOfChannels,
+ sourceSampleRate,
+ );
+ }
+
const combined = combineAudioDataAndClosePrevious(audioDataArray);
const resampled = resamplePcmS16AudioData({
targetSampleRate: sampleRate,
diff --git a/packages/media/src/test/extract-audio-resampling.test.ts b/packages/media/src/test/extract-audio-resampling.test.ts
index b2a4abe251..607082e90c 100644
--- a/packages/media/src/test/extract-audio-resampling.test.ts
+++ b/packages/media/src/test/extract-audio-resampling.test.ts
@@ -63,6 +63,35 @@ test('resamples Mediabunny AudioSamples as one continuous frame', async () => {
expect(maximumError).toBeLessThanOrEqual(2);
});
+test('pads the final silent tail of a looped audio frame', async () => {
+ const src = new URL('../../../remotion-media/ding.wav', import.meta.url).href;
+ const extracted = await extractAudio({
+ sampleRate: TARGET_SAMPLE_RATE,
+ audioStreamIndex: 0,
+ // Source frame 40 at 46x crosses the loop at frame 43. The WAV ends at
+ // frame 42, leaving silence at the end of both loop passes.
+ timeInSeconds: 42 / FPS,
+ durationInSeconds: 1 / FPS,
+ playbackRate: 46,
+ fps: FPS,
+ logLevel: 'info',
+ loop: true,
+ src,
+ trimBefore: 0,
+ trimAfter: 43,
+ maxCacheSize: getMaxVideoCacheSize('info'),
+ credentials: undefined,
+ mediaCache: globalMediaCache,
+ });
+
+ assert(typeof extracted === 'object' && extracted.data);
+ expect(extracted.data.numberOfFrames).toBe(TARGET_SAMPLE_RATE / FPS);
+ expect(extracted.data.data.some((sample) => sample !== 0)).toBe(true);
+ expect(Array.from(extracted.data.data.slice(-20))).toEqual(
+ new Array(20).fill(0),
+ );
+});
+
test('downmixes 5.1 audio without overflowing and excludes the LFE channel', () => {
const sourceChannels = new Int16Array([
32767, 32767, 32767, -32768, 32767, 32767, 0, 0, 0, 32767, 32767, -32768,