Add limited edit list support

dev/convert.html

@@ -28,6 +28,9 @@
 				format: outputFormat,
 				target
 			}),
+			audio: {
+				//discard: true
+			},
 			/*
 			video: {
 				discard: true,
@@ -43,6 +46,7 @@
 				codec: 'aac',
 			},
 			*/
+			/*
 			video: {
 				codec: 'avc',
 			},
@@ -50,9 +54,10 @@
 				codec: 'mp3',
 				bitrate: 320000
 			},
+			*/
 			trim: {
 				start: 0,
-				end: 60
+				end: 20
 			},
 			onProgress: (event) => {
 				progress.value = event.completion;

src/conversion.ts

@@ -341,7 +341,8 @@ class Conversion {
 			height = ceilToMultipleOfTwo(this.options.video.height);
 		}
 
-		const needsReencode = !!this.options.video?.forceReencode || this.startTimestamp > 0;
+		const firstTimestamp = await track.getFirstTimestamp();
+		const needsReencode = !!this.options.video?.forceReencode || this.startTimestamp > 0 || firstTimestamp < 0;
 		const needsResize = width !== originalWidth || height !== originalHeight;
 
 		let videoCodecs = this.output.format.getSupportedVideoCodecs();
@@ -513,11 +514,14 @@ class Conversion {
 		const originalNumberOfChannels = track.numberOfChannels;
 		const originalSampleRate = track.sampleRate;
 
+		const firstTimestamp = await track.getFirstTimestamp();
+
 		let numberOfChannels = this.options.audio?.numberOfChannels ?? originalNumberOfChannels;
 		let sampleRate = this.options.audio?.sampleRate ?? originalSampleRate;
 		let needsResample = numberOfChannels !== originalNumberOfChannels
 			|| sampleRate !== originalSampleRate
-			|| this.startTimestamp > 0;
+			|| this.startTimestamp > 0
+			|| firstTimestamp < 0;
 
 		let audioCodecs = this.output.format.getSupportedAudioCodecs();
 		if (
@@ -671,13 +675,12 @@ class Conversion {
 				await track.computeDuration() - this.startTimestamp,
 				this.endTimestamp - this.startTimestamp,
 			);
-			const totalFrameCount = Math.round(trackDuration * FALLBACK_SAMPLE_RATE);
-
-			const MAX_CHUNK_LENGTH = 5 * FALLBACK_SAMPLE_RATE;
+			const totalFrameCount = Math.round(trackDuration * targetSampleRate);
+			const maxChunkLength = 5 * targetSampleRate;
 
 			let currentContextStartFrame = 0;
 			let currentContext: OfflineAudioContext | null = new OfflineAudioContext({
-				length: Math.min(totalFrameCount - currentContextStartFrame, MAX_CHUNK_LENGTH),
+				length: Math.min(totalFrameCount - currentContextStartFrame, maxChunkLength),
 				numberOfChannels: targetNumberOfChannels,
 				sampleRate: targetSampleRate,
 			});
@@ -719,7 +722,7 @@ class Conversion {
 
 						const newLength = Math.min(
 							totalFrameCount - currentContextStartFrame,
-							MAX_CHUNK_LENGTH,
+							maxChunkLength,
 						);
 						currentContext = newLength > 0
 							? new OfflineAudioContext({

src/isobmff/isobmff-demuxer.ts

@@ -57,6 +57,7 @@ type InternalTrack = {
 	fragmentLookupTable: FragmentLookupTableEntry[] | null;
 	currentFragmentState: FragmentTrackState | null;
 	fragments: Fragment[];
+	editListOffset: number;
 } & ({
 	info: null;
 } | {
@@ -547,7 +548,8 @@ export class IsobmffDemuxer extends Demuxer {
 			case 'mdia':
 			case 'minf':
 			case 'dinf':
-			case 'mfra': {
+			case 'mfra':
+			case 'edts': {
 				this.readContiguousBoxes(boxInfo.contentSize);
 			}; break;
 
@@ -582,6 +584,7 @@ export class IsobmffDemuxer extends Demuxer {
 					fragmentLookupTable: null,
 					currentFragmentState: null,
 					fragments: [],
+					editListOffset: 0,
 				} satisfies InternalTrack as InternalTrack;
 				this.currentTrack = track;
 
@@ -654,6 +657,47 @@ export class IsobmffDemuxer extends Demuxer {
 				}
 			}; break;
 
+			case 'elst': {
+				const track = this.currentTrack;
+				assert(track);
+
+				const version = this.metadataReader.readU8();
+				this.metadataReader.pos += 3; // Flags
+
+				let relevantEntryFound = false;
+
+				const entryCount = this.metadataReader.readU32();
+				for (let i = 0; i < entryCount; i++) {
+					const segmentDuration = version === 1
+						? this.metadataReader.readU64()
+						: this.metadataReader.readU32();
+					const mediaTime = version === 1
+						? this.metadataReader.readI64()
+						: this.metadataReader.readI32();
+					const mediaRate = this.metadataReader.readFixed_16_16();
+
+					if (segmentDuration === 0) {
+						// Don't care
+						continue;
+					}
+
+					if (relevantEntryFound) {
+						throw new Error('Unsupported edit list: multiple edits are not supported.');
+					}
+
+					if (mediaTime === -1) {
+						throw new Error('Unsupported edit list: no empty edits allowed.');
+					}
+
+					if (mediaRate !== 1) {
+						throw new Error('Unsupported edit list: media rate must be 1.');
+					}
+
+					track.editListOffset = mediaTime;
+					relevantEntryFound = true;
+				}
+			}; break;
+
 			case 'mdhd': {
 				const track = this.currentTrack;
 				assert(track);
@@ -1796,15 +1840,15 @@ abstract class IsobmffTrackBacking<
 		return this.fetchSampleForSampleIndex(0, options);
 	}
 
-	private intoTimescale(timestamp: number) {
+	private mapTimestampIntoTimescale(timestamp: number) {
 		// Do a little rounding to catch cases where the result is very close to an integer. If it is, it's likely
 		// that the number was originally an integer divided by the timescale. For stability, it's best
 		// to return the integer in this case.
-		return roundToPrecision(timestamp * this.internalTrack.timescale, 14);
+		return roundToPrecision(timestamp * this.internalTrack.timescale, 14) + this.internalTrack.editListOffset;
 	}
 
 	async getSample(timestamp: number, options: SampleRetrievalOptions) {
-		const timestampInTimescale = this.intoTimescale(timestamp);
+		const timestampInTimescale = this.mapTimestampIntoTimescale(timestamp);
 
 		if (this.internalTrack.demuxer.isFragmented) {
 			return this.performFragmentedLookup(
@@ -1890,7 +1934,7 @@ abstract class IsobmffTrackBacking<
 	}
 
 	async getKeySample(timestamp: number, options: SampleRetrievalOptions) {
-		const timestampInTimescale = this.intoTimescale(timestamp);
+		const timestampInTimescale = this.mapTimestampIntoTimescale(timestamp);
 
 		if (this.internalTrack.demuxer.isFragmented) {
 			return this.performFragmentedLookup(
@@ -2014,7 +2058,8 @@ abstract class IsobmffTrackBacking<
 			data = this.internalTrack.demuxer.chunkReader.readBytes(sampleInfo.sampleSize);
 		}
 
-		const timestamp = sampleInfo.presentationTimestamp / this.internalTrack.timescale;
+		const timestamp = (sampleInfo.presentationTimestamp - this.internalTrack.editListOffset)
+			/ this.internalTrack.timescale;
 		const duration = sampleInfo.duration / this.internalTrack.timescale;
 		const sample = this.createSample(
 			data,
@@ -2049,7 +2094,8 @@ abstract class IsobmffTrackBacking<
 			data = this.internalTrack.demuxer.chunkReader.readBytes(fragmentSample.byteSize);
 		}
 
-		const timestamp = fragmentSample.presentationTimestamp / this.internalTrack.timescale;
+		const timestamp = (fragmentSample.presentationTimestamp - this.internalTrack.editListOffset)
+			/ this.internalTrack.timescale;
 		const duration = fragmentSample.duration / this.internalTrack.timescale;
 		const sample = this.createSample(
 			data,

todo.txt

@@ -1,4 +1,3 @@
-- Throw if edit list detected
 - A stream source?? Or like a callback-driven source
 - onHeader, etc callbacks for Matroska
 - https://github.com/Vanilagy/mp4-muxer/issues/83 tell him it's possible now