Implement Ogg demuxer

src/codec.ts

@@ -967,6 +967,27 @@ export const parseAacAudioSpecificConfig = (bytes: Uint8Array | null) => {
 	};
 };
 
+// From https://datatracker.ietf.org/doc/html/rfc6716, in 48 kHz samples
+const OPUS_FRAME_DURATION_TABLE = [
+	480, 960, 1920, 2880,
+	480, 960, 1920, 2880,
+	480, 960, 1920, 2880,
+	480, 960,
+	480, 960,
+	120, 240, 480, 960,
+	120, 240, 480, 960,
+	120, 240, 480, 960,
+	120, 240, 480, 960,
+];
+
+export const parseOpusTocByte = (packet: Uint8Array) => {
+	const config = packet[0]! >> 3;
+
+	return {
+		durationInSamples: OPUS_FRAME_DURATION_TABLE[config]!,
+	};
+};
+
 const PCM_CODEC_REGEX = /^pcm-([usf])(\d+)+(be)?$/;
 export const parsePcmCodec = (codec: PcmAudioCodec) => {
 	assert(PCM_AUDIO_CODECS.includes(codec));

src/index.ts

@@ -77,6 +77,7 @@ export {
 	WebMInputFormat,
 	Mp3InputFormat,
 	WaveInputFormat,
+	OggInputFormat,
 	ALL_FORMATS,
 	MP4,
 	QTFF,
@@ -84,6 +85,7 @@ export {
 	WEBM,
 	MP3,
 	WAVE,
+	OGG,
 } from './input-format';
 export { Input, InputOptions } from './input';
 export { InputTrack, InputVideoTrack, InputAudioTrack, SampleStats } from './input-track';

src/input-format.ts

@@ -6,6 +6,8 @@ import { EBMLId, EBMLReader } from './matroska/ebml';
 import { MatroskaDemuxer } from './matroska/matroska-demuxer';
 import { Mp3Demuxer } from './mp3/mp3-demuxer';
 import { Mp3Reader } from './mp3/mp3-reader';
+import { OggDemuxer } from './ogg/ogg-demuxer';
+import { OggReader } from './ogg/ogg-reader';
 import { RiffReader } from './wave/riff-reader';
 import { WaveDemuxer } from './wave/wave-demuxer';
 
@@ -196,9 +198,34 @@ export class Mp3InputFormat extends InputFormat {
 			mp3Reader.pos += id3Tag.size;
 		}
 
+		const framesStartPos = mp3Reader.pos;
 		await mp3Reader.reader.loadRange(mp3Reader.pos, mp3Reader.pos + 4096);
 
-		return mp3Reader.readNextFrameHeader(mp3Reader.pos + 4096) !== null;
+		const firstHeader = mp3Reader.readNextFrameHeader(framesStartPos + 4096);
+		if (!firstHeader) {
+			return false;
+		}
+
+		if (id3Tag) {
+			// If there was an ID3 tag at the start, we can be pretty sure this is MP3 by now
+			return true;
+		}
+
+		// Fine, we found one frame header, but we're still not entirely sure this is MP3. Let's check if we can find
+		// another header nearby:
+		mp3Reader.pos = firstHeader.startPos + firstHeader.totalSize;
+		const secondHeader = mp3Reader.readNextFrameHeader(framesStartPos + 4096);
+		if (!secondHeader) {
+			return false;
+		}
+
+		// In a well-formed MP3 file, we'd expect these two frames to share some similarities:
+		if (firstHeader.channel !== secondHeader.channel || firstHeader.sampleRate !== secondHeader.sampleRate) {
+			return false;
+		}
+
+		// We have found two matching MP3 frames, a strong indicator that this is an MP3 file
+		return true;
 	}
 
 	/** @internal */
@@ -248,6 +275,32 @@ export class WaveInputFormat extends InputFormat {
 	}
 }
 
+/** @public */
+export class OggInputFormat extends InputFormat {
+	async _canReadInput(input: Input) {
+		const sourceSize = await input._mainReader.source._getSize();
+		if (sourceSize < 4) {
+			return false;
+		}
+
+		const oggReader = new OggReader(input._mainReader);
+		return oggReader.readAscii(4) === 'OggS';
+	}
+
+	/** @internal */
+	_createDemuxer(input: Input) {
+		return new OggDemuxer(input);
+	}
+
+	getName() {
+		return 'Ogg';
+	}
+
+	getMimeType() {
+		return 'application/ogg';
+	}
+}
+
 /** @public */
 export const MP4 = new Mp4InputFormat();
 /** @public */
@@ -260,6 +313,8 @@ export const WEBM = new WebMInputFormat();
 export const MP3 = new Mp3InputFormat();
 /** @public */
 export const WAVE = new WaveInputFormat();
+/** @public */
+export const OGG = new OggInputFormat();
 
 /** @public */
-export const ALL_FORMATS: InputFormat[] = [MP4, QTFF, MATROSKA, WEBM, MP3, WAVE];
+export const ALL_FORMATS: InputFormat[] = [MP4, QTFF, MATROSKA, WEBM, WAVE, OGG, MP3];

src/isobmff/isobmff-demuxer.ts

@@ -37,6 +37,7 @@ import {
 	UNDETERMINED_LANGUAGE,
 	TransformationMatrix,
 	extractRotationFromMatrix,
+	roundToPrecision,
 } from '../misc';
 import { Reader } from '../reader';
 import { EncodedAudioSample, EncodedVideoSample, PLACEHOLDER_DATA, SampleType } from '../sample';
@@ -1792,16 +1793,10 @@ abstract class IsobmffTrackBacking<
 	}
 
 	private intoTimescale(timestamp: number) {
-		const result = timestamp * this.internalTrack.timescale;
+		// Do a little rounding to catch cases where the result is very close to an integer. If it is, it's likely
-		const rounded = Math.round(result);
+		// that the number was originally an integer divided by the timescale. For stability, it's best
-
+		// to return the integer in this case.
-		if (Math.abs(1 - (result / rounded)) < 10 * Number.EPSILON) {
+		return roundToPrecision(timestamp * this.internalTrack.timescale, 14);
-			// The result is very close to an integer, meaning the number likely originated by an integer being divided
-			// by the timescale. For stability, it's best to return the integer in this case.
-			return rounded;
-		}
-
-		return result;
 	}
 
 	async getSample(timestamp: number, options: SampleRetrievalOptions) {

src/matroska/matroska-demuxer.ts

@@ -29,6 +29,7 @@ import {
 	last,
 	MATRIX_COEFFICIENTS_MAP_INVERSE,
 	Rotation,
+	roundToPrecision,
 	TRANSFER_CHARACTERISTICS_MAP_INVERSE,
 	UNDETERMINED_LANGUAGE,
 } from '../misc';
@@ -1043,16 +1044,10 @@ abstract class MatroskaTrackBacking<
 	}
 
 	private intoTimescale(timestamp: number) {
-		const result = timestamp * this.internalTrack.segment.timestampFactor;
+		// Do a little rounding to catch cases where the result is very close to an integer. If it is, it's likely
-		const rounded = Math.round(result);
+		// that the number was originally an integer divided by the timescale. For stability, it's best
-
+		// to return the integer in this case.
-		if (Math.abs(1 - (result / rounded)) < 10 * Number.EPSILON) {
+		return roundToPrecision(timestamp * this.internalTrack.segment.timestampFactor, 14);
-			// The result is very close to an integer, meaning the number likely originated by an integer being divided
-			// by the timestamp factor. For stability, it's best to return the integer in this case.
-			return rounded;
-		}
-
-		return result;
 	}
 
 	async getSample(timestamp: number, options: SampleRetrievalOptions) {

src/misc.ts

@@ -237,6 +237,16 @@ export const removeItem = <T>(arr: T[], item: T) => {
 	}
 };
 
+export const findLast = <T>(arr: T[], predicate: (x: T) => boolean) => {
+	for (let i = arr.length - 1; i >= 0; i--) {
+		if (predicate(arr[i]!)) {
+			return arr[i];
+		}
+	}
+
+	return undefined;
+};
+
 export const findLastIndex = <T>(arr: T[], predicate: (x: T) => boolean) => {
 	for (let i = arr.length - 1; i >= 0; i--) {
 		if (predicate(arr[i]!)) {
@@ -367,3 +377,17 @@ export const setVideoFrameTiming = (frame: VideoFrame, timing: {
 
 	return clone;
 };
+
+export const roundToPrecision = (value: number, digits: number) => {
+	const factor = 10 ** digits;
+	return Math.round(value * factor) / factor;
+};
+
+export const ilog = (x: number) => {
+	let ret = 0;
+	while (x) {
+		ret++;
+		x >>= 1;
+	}
+	return ret;
+};

src/ogg/ogg-demuxer.ts

@@ -0,0 +1,986 @@
+import { AudioCodec, parseOpusTocByte } from '../codec';
+import { Demuxer } from '../demuxer';
+import { Input } from '../input';
+import { InputAudioTrack, InputAudioTrackBacking } from '../input-track';
+import { SampleRetrievalOptions } from '../media-sink';
+import { assert, findLast, ilog, roundToPrecision, toDataView, UNDETERMINED_LANGUAGE } from '../misc';
+import { Reader } from '../reader';
+import { EncodedAudioSample, PLACEHOLDER_DATA } from '../sample';
+import { computeOggPageCrc } from './ogg-misc';
+import { MAX_PAGE_HEADER_SIZE, MAX_PAGE_SIZE, MIN_PAGE_HEADER_SIZE, OggReader, Page } from './ogg-reader';
+import { parseModesFromSetupPacket } from './vorbis';
+
+type LogicalBitstream = {
+	serialNumber: number;
+	bosPage: Page;
+	codec: AudioCodec | null;
+	description: Uint8Array | null;
+	numberOfChannels: number;
+	sampleRate: number;
+	vorbisInfo: {
+		blocksizes: number[];
+		modeBlockflags: number[];
+	} | null;
+	opusInfo: {
+		preSkip: number;
+	} | null;
+
+	lastMetadataPacket: Packet | null;
+};
+
+type Packet = {
+	data: Uint8Array;
+	endPage: Page;
+	endSegmentIndex: number;
+};
+
+export class OggDemuxer extends Demuxer {
+	reader: OggReader;
+
+	metadataPromise: Promise<void> | null = null;
+	fileSize: number | null = null;
+	bitstreams: LogicalBitstream[] = [];
+	tracks: InputAudioTrack[] = [];
+
+	constructor(input: Input) {
+		super(input);
+
+		// We don't need a persistent metadata reader as we read all metadata once at the start and then never again
+		this.reader = new OggReader(new Reader(input._source, 64 * 2 ** 20));
+	}
+
+	async readMetadata() {
+		return this.metadataPromise ??= (async () => {
+			this.fileSize = await this.input._source._getSize();
+
+			while (this.reader.pos < this.fileSize - MIN_PAGE_HEADER_SIZE) {
+				await this.reader.reader.loadRange(
+					this.reader.pos,
+					this.reader.pos + MAX_PAGE_HEADER_SIZE,
+				);
+
+				const page = this.reader.readPageHeader();
+				if (!page) {
+					break;
+				}
+
+				const isBos = !!(page.headerType & 0x02);
+				if (!isBos) {
+					// All bos pages for all bitstreams are required to be at the start, so if the page is not bos then
+					// we know we've seen all bitstreams (minus chaining)
+					break;
+				}
+
+				this.bitstreams.push({
+					serialNumber: page.serialNumber,
+					bosPage: page,
+					codec: null,
+					description: null,
+					numberOfChannels: -1,
+					sampleRate: -1,
+					vorbisInfo: null,
+					opusInfo: null,
+					lastMetadataPacket: null,
+				});
+
+				this.reader.pos = page.headerStartPos + page.totalSize;
+			}
+
+			for (const bitstream of this.bitstreams) {
+				const firstPacket = await this.readPacket(this.reader, bitstream.bosPage, 0);
+				if (!firstPacket) {
+					continue;
+				}
+
+				if (
+					// Check for Vorbis
+					firstPacket.data.byteLength >= 7
+					&& firstPacket.data[0] === 0x01 // Packet type 1 = identification header
+					&& firstPacket.data[1] === 0x76 // 'v'
+					&& firstPacket.data[2] === 0x6f // 'o'
+					&& firstPacket.data[3] === 0x72 // 'r'
+					&& firstPacket.data[4] === 0x62 // 'b'
+					&& firstPacket.data[5] === 0x69 // 'i'
+					&& firstPacket.data[6] === 0x73 // 's'
+				) {
+					await this.readVorbisMetadata(firstPacket, bitstream);
+				} else if (
+					// Check for Opus
+					firstPacket.data.byteLength >= 8
+					&& firstPacket.data[0] === 0x4f // 'O'
+					&& firstPacket.data[1] === 0x70 // 'p'
+					&& firstPacket.data[2] === 0x75 // 'u'
+					&& firstPacket.data[3] === 0x73 // 's'
+					&& firstPacket.data[4] === 0x48 // 'H'
+					&& firstPacket.data[5] === 0x65 // 'e'
+					&& firstPacket.data[6] === 0x61 // 'a'
+					&& firstPacket.data[7] === 0x64 // 'd'
+				) {
+					await this.readOpusMetadata(firstPacket, bitstream);
+				}
+
+				if (bitstream.codec !== null) {
+					this.tracks.push(new InputAudioTrack(new OggAudioTrackBacking(bitstream, this)));
+				}
+			}
+		})();
+	}
+
+	async readVorbisMetadata(firstPacket: Packet, bitstream: LogicalBitstream) {
+		let nextPacketPosition = await this.findNextPacketStart(this.reader, firstPacket);
+		if (!nextPacketPosition) {
+			return;
+		}
+
+		const secondPacket = await this.readPacket(
+			this.reader,
+			nextPacketPosition.startPage,
+			nextPacketPosition.startSegmentIndex,
+		);
+		if (!secondPacket) {
+			return;
+		}
+
+		nextPacketPosition = await this.findNextPacketStart(this.reader, secondPacket);
+		if (!nextPacketPosition) {
+			return;
+		}
+
+		const thirdPacket = await this.readPacket(
+			this.reader,
+			nextPacketPosition.startPage,
+			nextPacketPosition.startSegmentIndex,
+		);
+		if (!thirdPacket) {
+			return;
+		}
+
+		if (secondPacket.data[0] !== 0x03 || thirdPacket.data[0] !== 0x05) {
+			return;
+		}
+
+		const lacingValues: number[] = [];
+		const addBytesToSegmentTable = (bytes: number) => {
+			while (true) {
+				lacingValues.push(Math.min(255, bytes));
+
+				if (bytes < 255) {
+					break;
+				}
+
+				bytes -= 255;
+			}
+		};
+
+		addBytesToSegmentTable(firstPacket.data.length);
+		addBytesToSegmentTable(secondPacket.data.length);
+		// We don't add the last packet to the segment table, as it is assumed to be whatever bytes remain
+
+		const description = new Uint8Array(
+			1 + lacingValues.length
+			+ firstPacket.data.length + secondPacket.data.length + thirdPacket.data.length,
+		);
+		description[0] = lacingValues.length;
+		description.set(
+			lacingValues, 1,
+		);
+		description.set(
+			firstPacket.data, 1 + lacingValues.length,
+		);
+		description.set(
+			secondPacket.data, 1 + lacingValues.length + firstPacket.data.length,
+		);
+		description.set(
+			thirdPacket.data, 1 + lacingValues.length + firstPacket.data.length + secondPacket.data.length,
+		);
+
+		bitstream.codec = 'vorbis';
+		bitstream.description = description;
+		bitstream.lastMetadataPacket = thirdPacket;
+
+		const view = toDataView(firstPacket.data);
+		bitstream.numberOfChannels = view.getUint8(11);
+		bitstream.sampleRate = view.getUint32(12, true);
+
+		const blockSizeByte = view.getUint8(28);
+		bitstream.vorbisInfo = {
+			blocksizes: [
+				1 << (blockSizeByte & 0xf),
+				1 << (blockSizeByte >> 4),
+			],
+			modeBlockflags: parseModesFromSetupPacket(thirdPacket.data).modeBlockflags,
+		};
+	}
+
+	async readOpusMetadata(firstPacket: Packet, bitstream: LogicalBitstream) {
+		// From https://datatracker.ietf.org/doc/html/rfc7845#section-5:
+		// "An Ogg Opus logical stream contains exactly two mandatory header packets: an identification header and a
+		// comment header."
+		const nextPacketPosition = await this.findNextPacketStart(this.reader, firstPacket);
+		if (!nextPacketPosition) {
+			return;
+		}
+
+		const secondPacket = await this.readPacket(
+			this.reader,
+			nextPacketPosition.startPage,
+			nextPacketPosition.startSegmentIndex,
+		);
+		if (!secondPacket) {
+			return;
+		}
+
+		// We don't make use of the comment header's data
+
+		bitstream.codec = 'opus';
+		bitstream.description = firstPacket.data;
+		bitstream.lastMetadataPacket = secondPacket;
+
+		const view = toDataView(firstPacket.data);
+		bitstream.numberOfChannels = view.getUint8(9);
+		bitstream.sampleRate = view.getUint32(12, true);
+
+		bitstream.opusInfo = {
+			preSkip: view.getUint16(10, true),
+		};
+	}
+
+	async readPacket(reader: OggReader, startPage: Page, startSegmentIndex: number): Promise<Packet | null> {
+		assert(startSegmentIndex < startPage.lacingValues.length);
+		assert(this.fileSize);
+
+		let startDataOffset = 0;
+		for (let i = 0; i < startSegmentIndex; i++) {
+			startDataOffset += startPage.lacingValues[i]!;
+		}
+
+		let currentPage: Page = startPage;
+		let currentDataOffset = startDataOffset;
+		let currentSegmentIndex = startSegmentIndex;
+
+		const chunks: Uint8Array[] = [];
+
+		outer:
+		while (true) {
+			// Load the entire page data
+			await reader.reader.loadRange(
+				currentPage.dataStartPos,
+				currentPage.dataStartPos + currentPage.dataSize,
+			);
+			reader.pos = currentPage.dataStartPos;
+			const pageData = reader.readBytes(currentPage.dataSize);
+
+			while (true) {
+				if (currentSegmentIndex === currentPage.lacingValues.length) {
+					chunks.push(pageData.subarray(startDataOffset, currentDataOffset));
+					break;
+				}
+
+				const lacingValue = currentPage.lacingValues[currentSegmentIndex]!;
+				currentDataOffset += lacingValue;
+
+				if (lacingValue < 255) {
+					chunks.push(pageData.subarray(startDataOffset, currentDataOffset));
+					break outer;
+				}
+
+				currentSegmentIndex++;
+			}
+
+			// The packet extends to the next page; let's find it
+			while (true) {
+				reader.pos = currentPage.headerStartPos + currentPage.totalSize;
+				if (reader.pos >= this.fileSize - MIN_PAGE_HEADER_SIZE) {
+					return null;
+				}
+
+				await reader.reader.loadRange(reader.pos, reader.pos + MAX_PAGE_HEADER_SIZE);
+				const nextPage = reader.readPageHeader();
+				if (!nextPage) {
+					return null;
+				}
+
+				currentPage = nextPage;
+				if (currentPage.serialNumber === startPage.serialNumber) {
+					break;
+				}
+			}
+
+			startDataOffset = 0;
+			currentDataOffset = 0;
+			currentSegmentIndex = 0;
+		}
+
+		const totalPacketSize = chunks.reduce((sum, chunk) => sum + chunk.length, 0);
+		const packetData = new Uint8Array(totalPacketSize);
+
+		let offset = 0;
+		for (let i = 0; i < chunks.length; i++) {
+			const chunk = chunks[i]!;
+			packetData.set(chunk, offset);
+			offset += chunk.length;
+		}
+
+		return {
+			data: packetData,
+			endPage: currentPage,
+			endSegmentIndex: currentSegmentIndex,
+		};
+	}
+
+	async findNextPacketStart(reader: OggReader, lastPacket: Packet) {
+		assert(this.fileSize !== null);
+
+		// If there's another segment in the same page, return it
+		if (lastPacket.endSegmentIndex < lastPacket.endPage.lacingValues.length - 1) {
+			return { startPage: lastPacket.endPage, startSegmentIndex: lastPacket.endSegmentIndex + 1 };
+		}
+
+		const isEos = !!(lastPacket.endPage.headerType & 0x04);
+		if (isEos) {
+			// The page is marked as the last page of the logical bitstream, so we won't find anything beyond it
+			return null;
+		}
+
+		// Otherwise, search for the next page belonging to the same bitstream
+		reader.pos = lastPacket.endPage.headerStartPos + lastPacket.endPage.totalSize;
+		while (true) {
+			if (reader.pos >= this.fileSize - MIN_PAGE_HEADER_SIZE) {
+				return null;
+			}
+
+			await reader.reader.loadRange(reader.pos, reader.pos + MAX_PAGE_HEADER_SIZE);
+			const nextPage = reader.readPageHeader();
+			if (!nextPage) {
+				return null;
+			}
+
+			if (nextPage.serialNumber === lastPacket.endPage.serialNumber) {
+				return { startPage: nextPage, startSegmentIndex: 0 };
+			}
+
+			reader.pos = nextPage.headerStartPos + nextPage.totalSize;
+		}
+	}
+
+	async getMimeType() {
+		await this.readMetadata();
+
+		let string = 'audio/ogg';
+
+		if (this.tracks.length > 0) {
+			const codecMimeTypes = await Promise.all(this.tracks.map(x => x.getCodecMimeType()));
+			const uniqueCodecMimeTypes = [...new Set(codecMimeTypes.filter(Boolean))];
+
+			string += `; codecs="${uniqueCodecMimeTypes.join(', ')}"`;
+		}
+
+		return string;
+	}
+
+	async getTracks() {
+		await this.readMetadata();
+		return this.tracks;
+	}
+
+	async computeDuration() {
+		const tracks = await this.getTracks();
+		const trackDurations = await Promise.all(tracks.map(x => x.computeDuration()));
+		return Math.max(0, ...trackDurations);
+	}
+}
+
+type SampleMetadata = {
+	packet: Packet;
+	timestampInSamples: number;
+	durationInSamples: number;
+	vorbisBlockSize: number | null;
+};
+
+class OggAudioTrackBacking implements InputAudioTrackBacking {
+	internalSampleRate: number;
+	sampleToMetadata = new WeakMap<EncodedAudioSample, SampleMetadata>();
+
+	constructor(public bitstream: LogicalBitstream, public demuxer: OggDemuxer) {
+		// Opus always uses 48000 for its internal calculations, even if the actual sample rate is different
+		this.internalSampleRate = bitstream.codec === 'opus' ? 48000 : bitstream.sampleRate;
+	}
+
+	getId() {
+		return this.bitstream.serialNumber;
+	}
+
+	async getNumberOfChannels() {
+		return this.bitstream.numberOfChannels;
+	}
+
+	async getSampleRate() {
+		return this.bitstream.sampleRate;
+	}
+
+	async getCodec() {
+		return this.bitstream.codec;
+	}
+
+	async getDecoderConfig(): Promise<AudioDecoderConfig | null> {
+		assert(this.bitstream.codec);
+
+		return {
+			codec: this.bitstream.codec,
+			numberOfChannels: this.bitstream.numberOfChannels,
+			sampleRate: this.bitstream.sampleRate,
+			description: this.bitstream.description ?? undefined,
+		};
+	}
+
+	async getLanguageCode() {
+		return UNDETERMINED_LANGUAGE;
+	}
+
+	async getFirstTimestamp() {
+		return 0;
+	}
+
+	async computeDuration() {
+		const lastSample = await this.getSample(Infinity, { metadataOnly: true });
+		return (lastSample?.timestamp ?? 0) + (lastSample?.duration ?? 0);
+	}
+
+	granulePositionToTimestampInSamples(granulePosition: number) {
+		if (this.bitstream.codec === 'opus') {
+			assert(this.bitstream.opusInfo);
+			return granulePosition - this.bitstream.opusInfo.preSkip;
+		}
+
+		return granulePosition;
+	}
+
+	createSampleFromPacket(
+		packet: Packet | null,
+		additional: {
+			timestampInSamples: number;
+			vorbisLastBlocksize: number | null;
+		},
+		options: SampleRetrievalOptions,
+	) {
+		if (!packet) {
+			return null;
+		}
+
+		let durationInSamples = 0;
+		let currentBlocksize: number | null = null;
+
+		if (packet.data.length > 0) {
+			// To know sample duration, we'll need to peak inside the packet
+
+			if (this.bitstream.codec === 'vorbis') {
+				assert(this.bitstream.vorbisInfo);
+
+				const vorbisModeCount = this.bitstream.vorbisInfo.modeBlockflags.length;
+				const bitCount = ilog(vorbisModeCount - 1);
+				const modeMask = ((1 << bitCount) - 1) << 1;
+				const modeNumber = (packet.data[0]! & modeMask) >> 1;
+
+				if (modeNumber >= this.bitstream.vorbisInfo.modeBlockflags.length) {
+					throw new Error('Invalid mode number.');
+				}
+
+				// In Vorbis, packet duration also depends on the blocksize of the previous packet
+				let prevBlocksize = additional.vorbisLastBlocksize;
+
+				const blockflag = this.bitstream.vorbisInfo.modeBlockflags[modeNumber]!;
+				currentBlocksize = this.bitstream.vorbisInfo.blocksizes[blockflag]!;
+
+				if (blockflag === 1) {
+					const prevMask = (modeMask | 0x1) + 1;
+					const flag = packet.data[0]! & prevMask ? 1 : 0;
+					prevBlocksize = this.bitstream.vorbisInfo.blocksizes[flag]!;
+				}
+
+				durationInSamples = prevBlocksize !== null
+					? (prevBlocksize + currentBlocksize) >> 2
+					: 0; // The first sample outputs no audio data and therefore has a duration of 0
+			} else if (this.bitstream.codec === 'opus') {
+				assert(this.bitstream.opusInfo);
+
+				const toc = parseOpusTocByte(packet.data);
+				durationInSamples = toc.durationInSamples;
+			}
+		}
+
+		const sample = new EncodedAudioSample(
+			options.metadataOnly ? PLACEHOLDER_DATA : packet.data,
+			'key',
+			Math.max(0, additional.timestampInSamples) / this.internalSampleRate,
+			durationInSamples / this.internalSampleRate,
+		);
+
+		this.sampleToMetadata.set(sample, {
+			packet,
+			timestampInSamples: additional.timestampInSamples,
+			durationInSamples,
+			vorbisBlockSize: currentBlocksize,
+		});
+		return sample;
+	}
+
+	async getFirstSample(options: SampleRetrievalOptions) {
+		assert(this.bitstream.lastMetadataPacket);
+		const packetPosition = await this.demuxer.findNextPacketStart(
+			this.demuxer.reader,
+			this.bitstream.lastMetadataPacket,
+		);
+		if (!packetPosition) {
+			return null;
+		}
+
+		let timestampInSamples = 0;
+		if (this.bitstream.codec === 'opus') {
+			assert(this.bitstream.opusInfo);
+			timestampInSamples -= this.bitstream.opusInfo.preSkip;
+		}
+
+		const packet = await this.demuxer.readPacket(
+			this.demuxer.reader,
+			packetPosition.startPage,
+			packetPosition.startSegmentIndex,
+		);
+
+		return this.createSampleFromPacket(
+			packet,
+			{
+				timestampInSamples,
+				vorbisLastBlocksize: null,
+			},
+			options,
+		);
+	}
+
+	async getNextSample(prevSample: EncodedAudioSample, options: SampleRetrievalOptions) {
+		const prevMetadata = this.sampleToMetadata.get(prevSample);
+		if (!prevMetadata) {
+			throw new Error('Sample was not created from this track.');
+		}
+
+		const packetPosition = await this.demuxer.findNextPacketStart(this.demuxer.reader, prevMetadata.packet);
+		if (!packetPosition) {
+			return null;
+		}
+
+		const timestampInSamples = prevMetadata.timestampInSamples + prevMetadata.durationInSamples;
+
+		const packet = await this.demuxer.readPacket(
+			this.demuxer.reader,
+			packetPosition.startPage,
+			packetPosition.startSegmentIndex,
+		);
+
+		return this.createSampleFromPacket(
+			packet,
+			{
+				timestampInSamples,
+				vorbisLastBlocksize: prevMetadata.vorbisBlockSize,
+			},
+			options,
+		);
+	}
+
+	async getSample(timestamp: number, options: SampleRetrievalOptions) {
+		assert(this.demuxer.fileSize !== null);
+
+		const timestampInSamples = roundToPrecision(timestamp * this.internalSampleRate, 14);
+		if (timestampInSamples === 0) {
+			// Fast path for timestamp 0 - avoids binary search when playing back from the start
+			return this.getFirstSample(options);
+		}
+		if (timestampInSamples < 0) {
+			// There's nothing here
+			return null;
+		}
+
+		const reader = this.demuxer.reader;
+
+		assert(this.bitstream.lastMetadataPacket);
+		const startPosition = await this.demuxer.findNextPacketStart(
+			reader,
+			this.bitstream.lastMetadataPacket,
+		);
+		if (!startPosition) {
+			return null;
+		}
+
+		let lowPage = startPosition.startPage;
+		let high = this.demuxer.fileSize;
+
+		const lowPages: Page[] = [lowPage];
+
+		// First, let's perform a binary serach (bisection search) on the file to find the approximate page where we'll
+		// find the sample. We want to find a page whose end sample position is less than or equal to the
+		// sample position we're searching for.
+
+		// Outer loop: Does the binary serach
+		outer:
+		while (lowPage.headerStartPos + lowPage.totalSize < high) {
+			const low = lowPage.headerStartPos;
+			const mid = Math.floor((low + high) / 2);
+
+			let searchStartPos = mid;
+
+			// Inner loop: Does a linear forward scan if the page cannot be found immediately
+			while (true) {
+				const until = Math.min(
+					searchStartPos + MAX_PAGE_SIZE,
+					high - MIN_PAGE_HEADER_SIZE,
+				);
+
+				await reader.reader.loadRange(searchStartPos, until);
+
+				reader.pos = searchStartPos;
+				const found = reader.findNextPageHeader(until);
+
+				if (!found) {
+					high = mid + MIN_PAGE_HEADER_SIZE;
+					continue outer;
+				}
+
+				await reader.reader.loadRange(reader.pos, reader.pos + MAX_PAGE_HEADER_SIZE);
+				const page = reader.readPageHeader();
+				assert(page);
+
+				let pageValid = false;
+				if (page.serialNumber === this.bitstream.serialNumber) {
+					// Serial numbers are basically random numbers, and the chance of finding a fake page with matching
+					// serial number is astronomically low, so we can be pretty sure this page is legit.
+					pageValid = true;
+				} else {
+					await reader.reader.loadRange(page.headerStartPos, page.headerStartPos + page.totalSize);
+
+					// Validate the page by checking checksum
+					reader.pos = page.headerStartPos;
+					const bytes = reader.readBytes(page.totalSize);
+					const crc = computeOggPageCrc(bytes);
+
+					pageValid = crc === page.checksum;
+				}
+
+				if (!pageValid) {
+					// Keep searching for a valid page
+					searchStartPos = page.headerStartPos + 4; // 'OggS' is 4 bytes
+					continue;
+				}
+
+				if (pageValid && page.serialNumber !== this.bitstream.serialNumber) {
+					// Page is valid but from a different bitstream, so keep searching forward until we find one
+					// belonging to the our bitstream
+					searchStartPos = page.headerStartPos + page.totalSize;
+					continue;
+				}
+
+				const isContinuationPage = page.granulePosition === -1;
+				if (isContinuationPage) {
+					// No sample ends on this page - keep looking
+					searchStartPos = page.headerStartPos + page.totalSize;
+					continue;
+				}
+
+				// The page is valid and belongs to our bitstream; let's check its granule position to see where we need
+				// to take the bisection search.
+				if (this.granulePositionToTimestampInSamples(page.granulePosition) > timestampInSamples) {
+					high = page.headerStartPos;
+				} else {
+					lowPage = page;
+					lowPages.push(page);
+				}
+
+				continue outer;
+			}
+		}
+
+		// Now we have the last page with a sample position <= the sample position we're looking for, but there might
+		// be multiple pages with the sample position, in which case we actually need to find the first of such pages.
+		// We'll do this in two steps: First, let's find the latest page we know with an earlier sample position, and
+		// then linear scan ourselves forward until we find the correct page.
+
+		let lowerPage = startPosition.startPage;
+		for (const otherLowPage of lowPages) {
+			if (otherLowPage.granulePosition === lowPage.granulePosition) {
+				break;
+			}
+
+			if (!lowerPage || otherLowPage.headerStartPos > lowerPage.headerStartPos) {
+				lowerPage = otherLowPage;
+			}
+		}
+
+		let currentPage: Page | null = lowerPage;
+		// Keep track of the pages we traversed, we need these later for backwards seeking
+		const previousPages: Page[] = [currentPage];
+
+		while (true) {
+			// This loop must terminate as we'll eventually reach lowPage
+			if (
+				currentPage.serialNumber === this.bitstream.serialNumber
+				&& currentPage.granulePosition === lowPage.granulePosition
+			) {
+				break;
+			}
+
+			reader.pos = currentPage.headerStartPos + currentPage.totalSize;
+			await reader.reader.loadRange(reader.pos, reader.pos + MAX_PAGE_HEADER_SIZE);
+
+			const nextPage = reader.readPageHeader();
+			assert(nextPage);
+
+			currentPage = nextPage;
+
+			if (currentPage.serialNumber === this.bitstream.serialNumber) {
+				previousPages.push(currentPage);
+			}
+		}
+
+		assert(currentPage.granulePosition !== -1);
+
+		let currentSegmentIndex: number | null = null;
+		let currentTimestampInSamples: number;
+		let currentTimestampIsCorrect: boolean;
+
+		// These indicate the end position of the packet that the granule position belongs to
+		let endPage = currentPage;
+		let endSegmentIndex = 0;
+
+		if (currentPage.headerStartPos === startPosition.startPage.headerStartPos) {
+			currentTimestampInSamples = this.granulePositionToTimestampInSamples(0);
+			currentTimestampIsCorrect = true;
+			currentSegmentIndex = 0;
+		} else {
+			currentTimestampInSamples = 0; // Placeholder value! We'll refine it once we can
+			currentTimestampIsCorrect = false;
+
+			// Find the segment index of the next packet
+			for (let i = currentPage.lacingValues.length - 1; i >= 0; i--) {
+				const value = currentPage.lacingValues[i]!;
+				if (value < 255) {
+					// We know the last packet ended at i, so the next one starts at i + 1
+					currentSegmentIndex = i + 1;
+					break;
+				}
+			}
+
+			// This must hold: Since this page has a granule position set, that means there must be a packet that ends
+			// in this page.
+			if (currentSegmentIndex === null) {
+				throw new Error('Invalid page with granule position: no packets end on this page.');
+			}
+
+			endSegmentIndex = currentSegmentIndex - 1;
+			const nextPosition = await this.demuxer.findNextPacketStart(reader, {
+				data: PLACEHOLDER_DATA,
+				endPage,
+				endSegmentIndex,
+			});
+
+			if (nextPosition) {
+				// Let's rewind a single step (packet) - this previous packet ensures that we'll correctly compute the
+				// duration for the packet we're looking for.
+				const endPosition = findPreviousPacketEndPosition(previousPages, currentPage, currentSegmentIndex);
+				assert(endPosition);
+
+				const startPosition = findPacketStartPosition(
+					previousPages, endPosition.page, endPosition.segmentIndex,
+				);
+				if (startPosition) {
+					currentPage = startPosition.page;
+					currentSegmentIndex = startPosition.segmentIndex;
+				}
+			} else {
+				// There is no next position, which means we're looking for the last sample in the bitstream. The
+				// granule position on the last page tends to be fucky, so let's instead start the search on the page
+				// before that. So let's loop until we find a packet that ends in a previous page.
+				while (true) {
+					const endPosition = findPreviousPacketEndPosition(previousPages, currentPage, currentSegmentIndex);
+					if (!endPosition) {
+						break;
+					}
+
+					const startPosition = findPacketStartPosition(
+						previousPages, endPosition.page, endPosition.segmentIndex,
+					);
+					if (!startPosition) {
+						break;
+					}
+
+					currentPage = startPosition.page;
+					currentSegmentIndex = startPosition.segmentIndex;
+
+					if (endPosition.page.headerStartPos !== endPage.headerStartPos) {
+						endPage = endPosition.page;
+						endSegmentIndex = endPosition.segmentIndex;
+						break;
+					}
+				}
+			}
+		}
+
+		let lastSample: EncodedAudioSample | null = null;
+		let lastSampleMetadata: SampleMetadata | null = null;
+
+		// Alright, now it's time for the final, granular seek: We keep iterating over packets until we've found the one
+		// with the correct timestamp - i.e., the last one with a timestamp <= the timestamp we're looking for.
+		while (currentPage !== null) {
+			assert(currentSegmentIndex !== null);
+
+			const packet = await this.demuxer.readPacket(reader, currentPage, currentSegmentIndex);
+			if (!packet) {
+				break;
+			}
+
+			// We might need to skip the packet if it's a metadata one
+			const skipPacket = currentPage.headerStartPos === startPosition.startPage.headerStartPos
+				&& currentSegmentIndex < startPosition.startSegmentIndex;
+
+			if (!skipPacket) {
+				let sample = this.createSampleFromPacket(
+					packet,
+					{
+						timestampInSamples: currentTimestampInSamples,
+						vorbisLastBlocksize: lastSampleMetadata?.vorbisBlockSize ?? null,
+					},
+					options,
+				);
+				assert(sample);
+
+				let sampleMetadata = this.sampleToMetadata.get(sample);
+				assert(sampleMetadata);
+
+				if (
+					!currentTimestampIsCorrect
+					&& packet.endPage.headerStartPos === endPage.headerStartPos
+					&& packet.endSegmentIndex === endSegmentIndex
+				) {
+					// We know this packet end timestamp can be derived from the page's granule position
+					currentTimestampInSamples = this.granulePositionToTimestampInSamples(currentPage.granulePosition);
+					currentTimestampIsCorrect = true;
+
+					// Let's backpatch the sample we just created with the correct timestamp
+					sample = this.createSampleFromPacket(
+						packet,
+						{
+							timestampInSamples: currentTimestampInSamples - sampleMetadata.durationInSamples,
+							vorbisLastBlocksize: lastSampleMetadata?.vorbisBlockSize ?? null,
+						},
+						options,
+					);
+					assert(sample);
+
+					sampleMetadata = this.sampleToMetadata.get(sample);
+					assert(sampleMetadata);
+				} else {
+					currentTimestampInSamples += sampleMetadata.durationInSamples;
+				}
+
+				lastSample = sample;
+				lastSampleMetadata = sampleMetadata;
+
+				if (
+					currentTimestampIsCorrect
+					&& (
+						// Next timestamp will be too late
+						Math.max(currentTimestampInSamples, 0) > timestampInSamples
+						// This timestamp already matches
+						|| Math.max(sampleMetadata.timestampInSamples, 0) === timestampInSamples
+					)
+				) {
+					break;
+				}
+			}
+
+			const nextPosition = await this.demuxer.findNextPacketStart(reader, packet);
+			if (!nextPosition) {
+				break;
+			}
+
+			currentPage = nextPosition.startPage;
+			currentSegmentIndex = nextPosition.startSegmentIndex;
+		}
+
+		return lastSample;
+	}
+
+	getKeySample(timestamp: number, options: SampleRetrievalOptions) {
+		return this.getSample(timestamp, options);
+	}
+
+	getNextKeySample(sample: EncodedAudioSample, options: SampleRetrievalOptions) {
+		return this.getNextSample(sample, options);
+	}
+}
+
+/** Finds the start position of a packet given its end position. */
+const findPacketStartPosition = (pageList: Page[], endPage: Page, endSegmentIndex: number) => {
+	let page = endPage;
+	let segmentIndex = endSegmentIndex;
+
+	outer:
+	while (true) {
+		segmentIndex--;
+
+		for (segmentIndex; segmentIndex >= 0; segmentIndex--) {
+			const lacingValue = page.lacingValues[segmentIndex]!;
+			if (lacingValue < 255) {
+				segmentIndex++; // We know the last sample starts here
+				break outer;
+			}
+		}
+
+		assert(segmentIndex === -1);
+
+		const pageStartsWithFreshPacket = !(page.headerType & 0x01);
+		if (pageStartsWithFreshPacket) {
+			// Fast exit: We know we don't need to look in the previous page
+			segmentIndex = 0;
+			break;
+		}
+
+		const previousPage = findLast(
+			pageList,
+			x => x.headerStartPos < page.headerStartPos,
+		);
+		if (!previousPage) {
+			return null;
+		}
+
+		page = previousPage;
+		segmentIndex = page.lacingValues.length;
+	}
+
+	assert(segmentIndex !== -1);
+
+	if (segmentIndex === page.lacingValues.length) {
+		// Wrap back around to the first segment of the next page
+		const nextPage = pageList[pageList.indexOf(page) + 1];
+		assert(nextPage);
+
+		page = nextPage;
+		segmentIndex = 0;
+	}
+
+	return { page, segmentIndex };
+};
+
+/** Finds the end position of a packet given the start position of the following packet. */
+const findPreviousPacketEndPosition = (pageList: Page[], startPage: Page, startSegmentIndex: number) => {
+	if (startSegmentIndex > 0) {
+		// Easy
+		return { page: startPage, segmentIndex: startSegmentIndex - 1 };
+	}
+
+	const previousPage = findLast(
+		pageList,
+		x => x.headerStartPos < startPage.headerStartPos,
+	);
+	if (!previousPage) {
+		return null;
+	}
+
+	return { page: previousPage, segmentIndex: previousPage.lacingValues.length - 1 };
+};

src/ogg/ogg-misc.ts

@@ -0,0 +1,32 @@
+import { toDataView } from '../misc';
+
+const OGG_CRC_POLYNOMIAL = 0x04c11db7;
+const OGG_CRC_TABLE = new Uint32Array(256);
+for (let n = 0; n < 256; n++) {
+	let crc = n << 24;
+
+	for (let k = 0; k < 8; k++) {
+		crc = (crc & 0x80000000)
+			? ((crc << 1) ^ OGG_CRC_POLYNOMIAL)
+			: (crc << 1);
+	}
+
+	OGG_CRC_TABLE[n] = (crc >>> 0) & 0xffffffff;
+}
+
+export const computeOggPageCrc = (bytes: Uint8Array) => {
+	const view = toDataView(bytes);
+
+	const originalChecksum = view.getUint32(22, true);
+	view.setUint32(22, 0, true); // Zero out checksum field
+
+	let crc = 0;
+	for (let i = 0; i < bytes.length; i++) {
+		const byte = bytes[i]!;
+		crc = ((crc << 8) ^ OGG_CRC_TABLE[(crc >>> 24) ^ byte]!) >>> 0;
+	}
+
+	view.setUint32(22, originalChecksum, true); // Restore checksum field
+
+	return crc;
+};

src/ogg/ogg-reader.ts

@@ -0,0 +1,135 @@
+import { Reader } from '../reader';
+
+const OGGS = 0x5367674f; // 'OggS'
+export const MIN_PAGE_HEADER_SIZE = 27;
+export const MAX_PAGE_HEADER_SIZE = 27 + 255;
+export const MAX_PAGE_SIZE = MAX_PAGE_HEADER_SIZE + 255 * 255;
+
+export type Page = {
+	headerStartPos: number;
+	totalSize: number;
+	dataStartPos: number;
+	dataSize: number;
+	headerType: number;
+	granulePosition: number;
+	serialNumber: number;
+	sequenceNumber: number;
+	checksum: number;
+	lacingValues: number[];
+};
+
+export class OggReader {
+	pos = 0;
+	constructor(public reader: Reader) {}
+
+	readBytes(length: number) {
+		const { view, offset } = this.reader.getViewAndOffset(this.pos, this.pos + length);
+		this.pos += length;
+
+		return new Uint8Array(view.buffer, offset, length);
+	}
+
+	readU8() {
+		const { view, offset } = this.reader.getViewAndOffset(this.pos, this.pos + 1);
+		this.pos += 1;
+
+		return view.getUint8(offset);
+	}
+
+	readU32() {
+		const { view, offset } = this.reader.getViewAndOffset(this.pos, this.pos + 4);
+		this.pos += 4;
+
+		return view.getUint32(offset, true);
+	}
+
+	readI32() {
+		const { view, offset } = this.reader.getViewAndOffset(this.pos, this.pos + 4);
+		this.pos += 4;
+
+		return view.getInt32(offset, true);
+	}
+
+	readI64() {
+		const low = this.readU32();
+		const high = this.readI32();
+		return high * 0x100000000 + low;
+	}
+
+	readAscii(length: number) {
+		const { view, offset } = this.reader.getViewAndOffset(this.pos, this.pos + length);
+		this.pos += length;
+
+		let str = '';
+		for (let i = 0; i < length; i++) {
+			str += String.fromCharCode(view.getUint8(offset + i));
+		}
+		return str;
+	}
+
+	readPageHeader(): Page | null {
+		const startPos = this.pos;
+
+		const capturePattern = this.readU32();
+		if (capturePattern !== OGGS) {
+			return null;
+		}
+
+		this.pos += 1; // Version
+		const headerType = this.readU8();
+		const granulePosition = this.readI64();
+		const serialNumber = this.readU32();
+		const sequenceNumber = this.readU32();
+		const checksum = this.readU32();
+
+		const numberPageSegments = this.readU8();
+		const lacingValues: number[] = [];
+
+		for (let i = 0; i < numberPageSegments; i++) {
+			lacingValues.push(this.readU8());
+		}
+
+		const headerSize = 27 + numberPageSegments;
+		const dataSize = lacingValues.reduce((a, b) => a + b, 0);
+		const totalSize = headerSize + dataSize;
+
+		return {
+			headerStartPos: startPos,
+			totalSize,
+			dataStartPos: startPos + headerSize,
+			dataSize,
+			headerType,
+			granulePosition,
+			serialNumber,
+			sequenceNumber,
+			checksum,
+			lacingValues,
+		};
+	}
+
+	findNextPageHeader(until: number) {
+		while (this.pos < until - (4 - 1)) { // Size of word minus 1
+			const word = this.readU32();
+			const firstByte = word & 0xff;
+			const secondByte = (word >>> 8) & 0xff;
+			const thirdByte = (word >>> 16) & 0xff;
+			const fourthByte = (word >>> 24) & 0xff;
+
+			const O = 0x4f; // 'O'
+			if (firstByte !== O && secondByte !== O && thirdByte !== O && fourthByte !== O) {
+				continue;
+			}
+
+			this.pos -= 4;
+
+			if (word === OGGS) {
+				// We have found the capture pattern
+				return true;
+			}
+
+			this.pos += 1;
+		}
+
+		return false;
+	}
+}

src/ogg/vorbis.ts

@@ -0,0 +1,122 @@
+import { readBits } from '../misc';
+
+class Bitstream {
+	bytes: Uint8Array;
+	pos: number;
+
+	constructor(bytes: Uint8Array) {
+		this.bytes = bytes;
+		this.pos = 0;
+	}
+
+	read(n: number): number {
+		const result = readBits(this.bytes, this.pos, this.pos + n);
+		this.pos += n;
+		return result;
+	}
+
+	skip(n: number): void {
+		this.pos += n;
+	}
+
+	getBitsLeft(): number {
+		return this.bytes.length * 8 - this.pos;
+	}
+
+	getBitCount(): number {
+		return this.pos;
+	}
+
+	clone(): Bitstream {
+		const clone = new Bitstream(this.bytes);
+		clone.pos = this.pos;
+		return clone;
+	}
+}
+
+// Based on vorbis_parser.c from FFmpeg.
+export const parseModesFromSetupPacket = (setupHeader: Uint8Array) => {
+	// Verify that this is a Setup header.
+	if (setupHeader.length < 7) {
+		throw new Error('Setup header is too short.');
+	}
+	if (setupHeader[0] !== 5) {
+		throw new Error('Wrong packet type in Setup header.');
+	}
+	const signature = String.fromCharCode(...setupHeader.slice(1, 7));
+	if (signature !== 'vorbis') {
+		throw new Error('Invalid packet signature in Setup header.');
+	}
+
+	// Reverse the entire buffer.
+	const bufSize = setupHeader.length;
+	const revBuffer = new Uint8Array(bufSize);
+	for (let i = 0; i < bufSize; i++) {
+		revBuffer[i] = setupHeader[bufSize - 1 - i]!;
+	}
+
+	// Initialize a Bitstream on the reversed buffer.
+	const bs = new Bitstream(revBuffer);
+
+	// --- Find the framing bit.
+	// In FFmpeg code, we scan until get_bits1() returns 1.
+	let gotFramingBit = 0;
+	while (bs.getBitsLeft() > 97) {
+		if (bs.read(1) === 1) {
+			gotFramingBit = bs.getBitCount();
+			break;
+		}
+	}
+	if (gotFramingBit === 0) {
+		throw new Error('Invalid Setup header: framing bit not found.');
+	}
+
+	// --- Search backwards for a valid mode header.
+	// We try to “guess” the number of modes by reading a fixed pattern.
+	let modeCount = 0;
+	let gotModeHeader = false;
+	let lastModeCount = 0;
+	while (bs.getBitsLeft() >= 97) {
+		const tempPos = bs.pos;
+		const a = bs.read(8);
+		const b = bs.read(16);
+		const c = bs.read(16);
+		// If a > 63 or b or c nonzero, assume we’ve gone too far.
+		if (a > 63 || b !== 0 || c !== 0) {
+			bs.pos = tempPos;
+			break;
+		}
+		bs.skip(1);
+		modeCount++;
+		if (modeCount > 64)
+			break;
+		const bsClone = bs.clone();
+		const candidate = bsClone.read(6) + 1;
+		if (candidate === modeCount) {
+			gotModeHeader = true;
+			lastModeCount = modeCount;
+		}
+	}
+	if (!gotModeHeader) {
+		throw new Error('Invalid Setup header: mode header not found.');
+	}
+	if (lastModeCount > 63) {
+		throw new Error(`Unsupported mode count: ${lastModeCount}.`);
+	}
+	const finalModeCount = lastModeCount;
+
+	// --- Reinitialize the bitstream.
+	bs.pos = 0;
+	// Skip the bits up to the found framing bit.
+	bs.skip(gotFramingBit);
+
+	// --- Now read, for each mode (in reverse order), 40 bits then one bit.
+	// That one bit is the mode blockflag.
+	const modeBlockflags = Array(finalModeCount).fill(0) as	number[];
+	for (let i = finalModeCount - 1; i >= 0; i--) {
+		bs.skip(40);
+		modeBlockflags[i] = bs.read(1);
+	}
+
+	return { modeBlockflags };
+};

src/sample.ts

@@ -11,7 +11,23 @@ export class EncodedVideoSample {
 		public readonly timestamp: number,
 		public readonly duration: number,
 		public readonly byteLength = data.byteLength,
-	) {}
+	) {
+		if (!(data instanceof Uint8Array)) {
+			throw new TypeError('data must be a Uint8Array.');
+		}
+		if (type !== 'key' && type !== 'delta') {
+			throw new TypeError('type must be either "key" or "delta".');
+		}
+		if (!Number.isFinite(timestamp)) {
+			throw new TypeError('timestamp must be a number.');
+		}
+		if (!Number.isFinite(duration) || duration < 0) {
+			throw new TypeError('duration must be a non-negative number.');
+		}
+		if (!Number.isInteger(byteLength) || byteLength < 0) {
+			throw new TypeError('byteLength must be a non-negative integer.');
+		}
+	}
 
 	get isMetadataOnly() {
 		return this.data === PLACEHOLDER_DATA;
@@ -103,7 +119,23 @@ export class EncodedAudioSample {
 		public readonly timestamp: number,
 		public readonly duration: number,
 		public readonly byteLength = data.byteLength,
-	) {}
+	) {
+		if (!(data instanceof Uint8Array)) {
+			throw new TypeError('data must be a Uint8Array.');
+		}
+		if (type !== 'key' && type !== 'delta') {
+			throw new TypeError('type must be either "key" or "delta".');
+		}
+		if (!Number.isFinite(timestamp)) {
+			throw new TypeError('timestamp must be a number.');
+		}
+		if (!Number.isFinite(duration) || duration < 0) {
+			throw new TypeError('duration must be a non-negative number.');
+		}
+		if (!Number.isInteger(byteLength) || byteLength < 0) {
+			throw new TypeError('byteLength must be a non-negative integer.');
+		}
+	}
 
 	get isMetadataOnly() {
 		return this.data === PLACEHOLDER_DATA;