Use decoder information for resampler sample rate, fix Opus sample rate to 48 kHz everywhere

Release

dev/convert.html

@@ -24,7 +24,7 @@
 			chunked: true,
 			chunkSize: 2**20
 		});
-		const outputFormat = new Mediabunny.Mp4OutputFormat({});
+		const outputFormat = new Mediabunny.WavOutputFormat({});
 
 		const button = document.createElement('button');
 		button.textContent = 'Cancel';
@@ -72,6 +72,9 @@
 			}),
 			output,
 			audio: {
+				codec: 'pcm-s16',
+				//sampleRate: 16000,
+				//numberOfChannels: 1,
 				//discard: true,
 				//codec: 'opus',
 				//bitrate: 128000,
@@ -106,7 +109,7 @@
 			*/
 			video: () => ({
 				//discard: true,
-				forceTranscode: true,
+				//forceTranscode: true,
 				//codec: 'avc',
 				//fit: 'contain',
 				//frameRate: 27.123,
@@ -129,7 +132,7 @@
 				//height: 100,
 			}),
 			trim: {
-				start: 1,
+				start: 0,
 				end: 10
 			},
 		});

dev/demux.html

@@ -14,10 +14,20 @@
 			source: new Mediabunny.BlobSource(file),
 		});
 
-		const videoTrack = await input.getPrimaryVideoTrack();
+		const audioTrack = await input.getPrimaryAudioTrack();
-		const sink = new Mediabunny.VideoSampleSink(videoTrack);
+		const sink = new Mediabunny.EncodedPacketSink(audioTrack);
 
-		console.log(await sink.getSample(await videoTrack.getFirstTimestamp()))
+		for await (const packet of sink.packets()) {
+			console.log(packet);
+		}
+
+		/*
+		const sink = new Mediabunny.EncodedPacketSink(videoTrack);
+
+		for await (const packet of sink.packets()) {
+			console.log(packet)
+		}
+		*/
 
 		/*
 		for await (const sample of sink.samples(0.99)) {

docs/guide/reading-media-files.md

@@ -60,6 +60,12 @@ await input.computeDuration(); // => 1905.4615
 ```
 More specifically, the duration is defined as the maximum end timestamp across all tracks.
 
+Mediabunny also lets you read descriptive metadata tags from media files, such as title, artist, or cover art:
+```ts
+await input.getMetadataTags(); // => MetadataTags
+```
+For more info, see [`MetadataTags`](../api/MetadataTags).
+
 ## Reading track metadata
 
 You can extract the list of all media tracks in the file like so:
@@ -76,12 +82,6 @@ await input.getPrimaryVideoTrack(); // => InputVideoTrack | null
 await input.getPrimaryAudioTrack(); // => InputAudioTrack | null
 ```
 
-Mediabunny also lets you read descriptive metadata tags from media files, such as title, artist, or cover art:
-```ts
-await input.getMetadataTags(); // => MetadataTags
-```
-For more info, see [`MetadataTags`](../api/MetadataTags).
-
 ::: info
 Subtitle tracks are currently not supported for reading.
 :::

docs/index.md

@@ -105,6 +105,7 @@ const sponsors = {
 		{ image: 'https://avatars.githubusercontent.com/u/9549394', name: 'studnitz', url: 'https://github.com/studnitz' },
 		{ image: 'https://avatars.githubusercontent.com/u/504909', name: 'Hirbod', url: 'https://github.com/hirbod' },
 		{ image: 'https://avatars.githubusercontent.com/u/30229596', name: 'Pablo Bonilla', url: 'https://github.com/devPablo' },
+		{ image: 'https://avatars.githubusercontent.com/u/38181164', name: 'wcw', url: 'https://github.com/asd55667' },
 		{ image: 'https://avatars.githubusercontent.com/u/63088713', name: 'taf2000', url: 'https://github.com/taf2000' },
 		{ image: 'https://avatars.githubusercontent.com/u/58149663', name: 'H7GhosT', url: 'https://github.com/H7GhosT' },
 		{ image: 'https://avatars.githubusercontent.com/u/91711202', name: 'ihasq', url: 'https://github.com/ihasq' },

examples/thumbnail-generation/thumbnail-generation.ts

@@ -96,6 +96,13 @@ const generateThumbnails = async (resource: File | string) => {
 				timestampElement.className
                     = 'absolute bottom-0 right-0 bg-black/30 text-white px-1 py-0.5 text-[11px] rounded-tl-lg';
 				container.append(timestampElement);
+			} else {
+				// Add something to indicate that the thumbnail is missing
+				const p = document.createElement('p');
+				p.textContent = '?';
+				p.className = 'absolute inset-0 flex items-center justify-center text-3xl opacity-50';
+
+				container.append(p);
 			}
 
 			i++;

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "mediabunny",
-  "version": "1.14.2",
+  "version": "1.14.4",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "mediabunny",
-      "version": "1.14.2",
+      "version": "1.14.4",
       "license": "MPL-2.0",
       "workspaces": [
         "packages/*"
@@ -7749,9 +7749,9 @@
       }
     },
     "node_modules/mediabunny": {
-      "version": "1.14.1",
+      "version": "1.14.3",
-      "resolved": "https://registry.npmjs.org/mediabunny/-/mediabunny-1.14.1.tgz",
+      "resolved": "https://registry.npmjs.org/mediabunny/-/mediabunny-1.14.3.tgz",
-      "integrity": "sha512-TjLg8GQGGsnGePcA6i0NItGe9a5G8WaleCGaXpb5V5okwuK5KpNKfcmTZXItnjfPLo7FvfEZI0NFp1lIR8Os7Q==",
+      "integrity": "sha512-kCvieRo6X1QDcdWLjn7o2BY/VCDeyU9nNGBVjOIOiWPoTtIekHR+viKAYaZafLEm0poBv+O2PwttL37PaSo/kA==",
       "license": "MPL-2.0",
       "peer": true,
       "workspaces": [
@@ -12242,7 +12242,7 @@
     },
     "packages/mp3-encoder": {
       "name": "@mediabunny/mp3-encoder",
-      "version": "1.14.2",
+      "version": "1.14.4",
       "license": "MPL-2.0",
       "devDependencies": {
         "@types/emscripten": "^1.40.1"

package.json

@@ -1,7 +1,7 @@
 {
   "name": "mediabunny",
   "author": "Vanilagy",
-  "version": "1.14.2",
+  "version": "1.14.4",
   "description": "Pure TypeScript media toolkit for reading, writing, and converting media files, directly in the browser.",
   "type": "module",
   "workspaces": [

packages/mp3-encoder/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@mediabunny/mp3-encoder",
   "author": "Vanilagy",
-  "version": "1.14.2",
+  "version": "1.14.4",
   "description": "MP3 encoder extension for Mediabunny, based on LAME.",
   "main": "./dist/bundles/mediabunny-mp3-encoder.mjs",
   "module": "./dist/bundles/mediabunny-mp3-encoder.mjs",

src/codec-data.ts

@@ -26,8 +26,27 @@ import { EncodedPacket, PacketType } from './packet';
 // Rec. ITU-T H.265
 // https://stackoverflow.com/questions/24884827
 
+export enum AvcNalUnitType {
+	IDR = 5,
+	SPS = 7,
+	PPS = 8,
+	SPS_EXT = 13,
+}
+
+export enum HevcNalUnitType {
+	RASL_N = 8,
+	RASL_R = 9,
+	BLA_W_LP = 16,
+	RSV_IRAP_VCL23 = 23,
+	VPS_NUT = 32,
+	SPS_NUT = 33,
+	PPS_NUT = 34,
+	PREFIX_SEI_NUT = 39,
+	SUFFIX_SEI_NUT = 40,
+}
+
 /** Finds all NAL units in an AVC packet in Annex B format. */
-const findNalUnitsInAnnexB = (packetData: Uint8Array) => {
+export const findNalUnitsInAnnexB = (packetData: Uint8Array) => {
 	const nalUnits: Uint8Array[] = [];
 	let i = 0;
 
@@ -184,6 +203,21 @@ export type AvcDecoderConfigurationRecord = {
 	sequenceParameterSetExt: Uint8Array[] | null;
 };
 
+export const extractAvcNalUnits = (packetData: Uint8Array, decoderConfig: VideoDecoderConfig) => {
+	if (decoderConfig.description) {
+		// Stream is length-prefixed. Let's extract the size of the length prefix from the decoder config
+
+		const bytes = toUint8Array(decoderConfig.description);
+		const lengthSizeMinusOne = bytes[4]! & 0b11;
+		const lengthSize = (lengthSizeMinusOne + 1) as 1 | 2 | 3 | 4;
+
+		return findNalUnitsInLengthPrefixed(packetData, lengthSize);
+	} else {
+		// Stream is in Annex B format
+		return findNalUnitsInAnnexB(packetData);
+	}
+};
+
 const extractNalUnitTypeForAvc = (data: Uint8Array) => {
 	return data[0]! & 0x1F;
 };
@@ -193,9 +227,9 @@ export const extractAvcDecoderConfigurationRecord = (packetData: Uint8Array) =>
 	try {
 		const nalUnits = findNalUnitsInAnnexB(packetData);
 
-		const spsUnits = nalUnits.filter(unit => extractNalUnitTypeForAvc(unit) === 7);
+		const spsUnits = nalUnits.filter(unit => extractNalUnitTypeForAvc(unit) === AvcNalUnitType.SPS);
-		const ppsUnits = nalUnits.filter(unit => extractNalUnitTypeForAvc(unit) === 8);
+		const ppsUnits = nalUnits.filter(unit => extractNalUnitTypeForAvc(unit) === AvcNalUnitType.PPS);
-		const spsExtUnits = nalUnits.filter(unit => extractNalUnitTypeForAvc(unit) === 13);
+		const spsExtUnits = nalUnits.filter(unit => extractNalUnitTypeForAvc(unit) === AvcNalUnitType.SPS_EXT);
 
 		if (spsUnits.length === 0) {
 			return null;
@@ -337,12 +371,6 @@ export const serializeAvcDecoderConfigurationRecord = (record: AvcDecoderConfigu
 	return new Uint8Array(bytes);
 };
 
-const NALU_TYPE_VPS = 32;
-const NALU_TYPE_SPS = 33;
-const NALU_TYPE_PPS = 34;
-const NALU_TYPE_SEI_PREFIX = 39;
-const NALU_TYPE_SEI_SUFFIX = 40;
-
 // Data specified in ISO 14496-15
 export type HevcDecoderConfigurationRecord = {
 	configurationVersion: number;
@@ -369,7 +397,22 @@ export type HevcDecoderConfigurationRecord = {
 	}[];
 };
 
-const extractNalUnitTypeForHevc = (data: Uint8Array) => {
+export const extractHevcNalUnits = (packetData: Uint8Array, decoderConfig: VideoDecoderConfig) => {
+	if (decoderConfig.description) {
+		// Stream is length-prefixed. Let's extract the size of the length prefix from the decoder config
+
+		const bytes = toUint8Array(decoderConfig.description);
+		const lengthSizeMinusOne = bytes[21]! & 0b11;
+		const lengthSize = (lengthSizeMinusOne + 1) as 1 | 2 | 3 | 4;
+
+		return findNalUnitsInLengthPrefixed(packetData, lengthSize);
+	} else {
+		// Stream is in Annex B format
+		return findNalUnitsInAnnexB(packetData);
+	}
+};
+
+export const extractNalUnitTypeForHevc = (data: Uint8Array) => {
 	return (data[0]! >> 1) & 0x3F;
 };
 
@@ -380,12 +423,12 @@ export const extractHevcDecoderConfigurationRecord = (
 	try {
 		const nalUnits = findNalUnitsInAnnexB(packetData);
 
-		const vpsUnits = nalUnits.filter(unit => extractNalUnitTypeForHevc(unit) === NALU_TYPE_VPS);
+		const vpsUnits = nalUnits.filter(unit => extractNalUnitTypeForHevc(unit) === HevcNalUnitType.VPS_NUT);
-		const spsUnits = nalUnits.filter(unit => extractNalUnitTypeForHevc(unit) === NALU_TYPE_SPS);
+		const spsUnits = nalUnits.filter(unit => extractNalUnitTypeForHevc(unit) === HevcNalUnitType.SPS_NUT);
-		const ppsUnits = nalUnits.filter(unit => extractNalUnitTypeForHevc(unit) === NALU_TYPE_PPS);
+		const ppsUnits = nalUnits.filter(unit => extractNalUnitTypeForHevc(unit) === HevcNalUnitType.PPS_NUT);
 		const seiUnits = nalUnits.filter(
-			unit => extractNalUnitTypeForHevc(unit) === NALU_TYPE_SEI_PREFIX
+			unit => extractNalUnitTypeForHevc(unit) === HevcNalUnitType.PREFIX_SEI_NUT
-				|| extractNalUnitTypeForHevc(unit) === NALU_TYPE_SEI_SUFFIX,
+				|| extractNalUnitTypeForHevc(unit) === HevcNalUnitType.SUFFIX_SEI_NUT,
 		);
 
 		if (spsUnits.length === 0 || ppsUnits.length === 0) return null;
@@ -521,7 +564,7 @@ export const extractHevcDecoderConfigurationRecord = (
 				? [
 						{
 							arrayCompleteness: 1,
-							nalUnitType: NALU_TYPE_VPS,
+							nalUnitType: HevcNalUnitType.VPS_NUT,
 							nalUnits: vpsUnits,
 						},
 					]
@@ -530,7 +573,7 @@ export const extractHevcDecoderConfigurationRecord = (
 				? [
 						{
 							arrayCompleteness: 1,
-							nalUnitType: NALU_TYPE_SPS,
+							nalUnitType: HevcNalUnitType.SPS_NUT,
 							nalUnits: spsUnits,
 						},
 					]
@@ -539,7 +582,7 @@ export const extractHevcDecoderConfigurationRecord = (
 				? [
 						{
 							arrayCompleteness: 1,
-							nalUnitType: NALU_TYPE_PPS,
+							nalUnitType: HevcNalUnitType.PPS_NUT,
 							nalUnits: ppsUnits,
 						},
 					]
@@ -1440,22 +1483,9 @@ export const determineVideoPacketType = async (
 			const decoderConfig = await videoTrack.getDecoderConfig();
 			assert(decoderConfig);
 
-			let nalUnits: Uint8Array[];
+			const nalUnits = extractAvcNalUnits(packet.data, decoderConfig);
+			const isKeyframe = nalUnits.some(x => extractNalUnitTypeForAvc(x) === AvcNalUnitType.IDR);
 
-			if (decoderConfig.description) {
-				// Stream is length-prefixed. Let's extract the size of the length prefix from the decoder config
-
-				const bytes = toUint8Array(decoderConfig.description);
-				const lengthSizeMinusOne = bytes[4]! & 0b11;
-				const lengthSize = (lengthSizeMinusOne + 1) as 1 | 2 | 3 | 4;
-
-				nalUnits = findNalUnitsInLengthPrefixed(packet.data, lengthSize);
-			} else {
-				// Stream is in Annex B format
-				nalUnits = findNalUnitsInAnnexB(packet.data);
-			}
-
-			const isKeyframe = nalUnits.some(x => extractNalUnitTypeForAvc(x) === 5);
 			return isKeyframe ? 'key' : 'delta';
 		};
 
@@ -1463,25 +1493,12 @@ export const determineVideoPacketType = async (
 			const decoderConfig = await videoTrack.getDecoderConfig();
 			assert(decoderConfig);
 
-			let nalUnits: Uint8Array[];
+			const nalUnits = extractHevcNalUnits(packet.data, decoderConfig);
-
-			if (decoderConfig.description) {
-				// Stream is length-prefixed. Let's extract the size of the length prefix from the decoder config
-
-				const bytes = toUint8Array(decoderConfig.description);
-				const lengthSizeMinusOne = bytes[21]! & 0b11;
-				const lengthSize = (lengthSizeMinusOne + 1) as 1 | 2 | 3 | 4;
-
-				nalUnits = findNalUnitsInLengthPrefixed(packet.data, lengthSize);
-			} else {
-				// Stream is in Annex B format
-				nalUnits = findNalUnitsInAnnexB(packet.data);
-			}
-
 			const isKeyframe = nalUnits.some((x) => {
 				const type = extractNalUnitTypeForHevc(x);
-				return 16 <= type && type <= 23;
+				return HevcNalUnitType.BLA_W_LP <= type && type <= HevcNalUnitType.RSV_IRAP_VCL23;
 			});
+
 			return isKeyframe ? 'key' : 'delta';
 		};
 

src/codec.ts

@@ -621,7 +621,7 @@ export const parseAacAudioSpecificConfig = (bytes: Uint8Array | null): AacAudioS
 	};
 };
 
-export const OPUS_INTERNAL_SAMPLE_RATE = 48000;
+export const OPUS_SAMPLE_RATE = 48_000;
 
 const PCM_CODEC_REGEX = /^pcm-([usf])(\d+)+(be)?$/;
 

src/conversion.ts

@@ -1125,8 +1125,6 @@ export class Conversion {
 			await this._started;
 
 			const resampler = new AudioResampler({
-				sourceNumberOfChannels: track.numberOfChannels,
-				sourceSampleRate: track.sampleRate,
 				targetNumberOfChannels,
 				targetSampleRate,
 				startTime: this._startTimestamp,
@@ -1246,9 +1244,9 @@ class TrackSynchronizer {
  * OfflineAudioContext.
  */
 export class AudioResampler {
-	sourceSampleRate: number;
+	sourceSampleRate: number | null = null;
 	targetSampleRate: number;
-	sourceNumberOfChannels: number;
+	sourceNumberOfChannels: number | null = null;
 	targetNumberOfChannels: number;
 	startTime: number;
 	endTime: number;
@@ -1262,20 +1260,16 @@ export class AudioResampler {
 	/** The highest index written to in the current buffer */
 	maxWrittenFrame: number;
 	channelMixer!: (sourceData: Float32Array, sourceFrameIndex: number, targetChannelIndex: number) => number;
-	tempSourceBuffer: Float32Array;
+	tempSourceBuffer!: Float32Array;
 
 	constructor(options: {
-		sourceSampleRate: number;
 		targetSampleRate: number;
-		sourceNumberOfChannels: number;
 		targetNumberOfChannels: number;
 		startTime: number;
 		endTime: number;
 		onSample: (sample: AudioSample) => Promise<void>;
 	}) {
-		this.sourceSampleRate = options.sourceSampleRate;
 		this.targetSampleRate = options.targetSampleRate;
-		this.sourceNumberOfChannels = options.sourceNumberOfChannels;
 		this.targetNumberOfChannels = options.targetNumberOfChannels;
 		this.startTime = options.startTime;
 		this.endTime = options.endTime;
@@ -1287,17 +1281,14 @@ export class AudioResampler {
 		this.outputBuffer = new Float32Array(this.bufferSizeInSamples);
 		this.bufferStartFrame = 0;
 		this.maxWrittenFrame = -1;
-
-		this.setupChannelMixer();
-
-		// Pre-allocate temporary buffer for source data
-		this.tempSourceBuffer = new Float32Array(this.sourceSampleRate * this.sourceNumberOfChannels);
 	}
 
 	/**
 	 * Sets up the channel mixer to handle up/downmixing in the case where input and output channel counts don't match.
 	 */
-	setupChannelMixer(): void {
+	doChannelMixerSetup(): void {
+		assert(this.sourceNumberOfChannels !== null);
+
 		const sourceNum = this.sourceNumberOfChannels;
 		const targetNum = this.targetNumberOfChannels;
 
@@ -1415,8 +1406,17 @@ export class AudioResampler {
 	}
 
 	async add(audioSample: AudioSample) {
-		if (!audioSample || audioSample._closed) {
+		if (this.sourceSampleRate === null) {
-			return;
+			// This is the first sample, so let's init the missing data. Initting the sample rate from the decoded
+			// sample is more reliable than using the file's metadata, because decoders are free to emit any sample rate
+			// they see fit.
+			this.sourceSampleRate = audioSample.sampleRate;
+			this.sourceNumberOfChannels = audioSample.numberOfChannels;
+
+			// Pre-allocate temporary buffer for source data
+			this.tempSourceBuffer = new Float32Array(this.sourceSampleRate * this.sourceNumberOfChannels);
+
+			this.doChannelMixerSetup();
 		}
 
 		const requiredSamples = audioSample.numberOfFrames * audioSample.numberOfChannels;

src/isobmff/isobmff-demuxer.ts

@@ -12,6 +12,7 @@ import {
 	extractAudioCodecString,
 	extractVideoCodecString,
 	MediaCodec,
+	OPUS_SAMPLE_RATE,
 	parseAacAudioSpecificConfig,
 	parsePcmCodec,
 	PCM_AUDIO_CODECS,
@@ -1049,6 +1050,10 @@ export class IsobmffDemuxer extends Demuxer {
 							}
 						}
 
+						if (track.info.codec === 'opus') {
+							sampleRate = OPUS_SAMPLE_RATE; // Always the same
+						}
+
 						track.info.numberOfChannels = channelCount;
 						track.info.sampleRate = sampleRate;
 
@@ -1421,7 +1426,7 @@ export class IsobmffDemuxer extends Demuxer {
 
 				track.info.codecDescription = description;
 				track.info.numberOfChannels = outputChannelCount;
-				track.info.sampleRate = inputSampleRate;
+				// Don't copy the input sample rate, irrelevant, and output sample rate is fixed
 			}; break;
 
 			case 'dfLa': { // Used for FLAC audio

src/matroska/matroska-demuxer.ts

@@ -18,6 +18,7 @@ import {
 	extractAudioCodecString,
 	extractVideoCodecString,
 	MediaCodec,
+	OPUS_SAMPLE_RATE,
 	VideoCodec,
 } from '../codec';
 import { Demuxer } from '../demuxer';
@@ -988,6 +989,7 @@ export class MatroskaDemuxer extends Demuxer {
 						} else if (codecIdWithoutSuffix === CODEC_STRING_MAP.opus) {
 							this.currentTrack.info.codec = 'opus';
 							this.currentTrack.info.codecDescription = this.currentTrack.codecPrivate;
+							this.currentTrack.info.sampleRate = OPUS_SAMPLE_RATE; // Always the same
 						} else if (codecIdWithoutSuffix === CODEC_STRING_MAP.vorbis) {
 							this.currentTrack.info.codec = 'vorbis';
 							this.currentTrack.info.codecDescription = this.currentTrack.codecPrivate;

src/matroska/matroska-muxer.ts

@@ -47,7 +47,7 @@ import {
 	parseSubtitleTimestamp,
 } from '../subtitles';
 import {
-	OPUS_INTERNAL_SAMPLE_RATE,
+	OPUS_SAMPLE_RATE,
 	PCM_AUDIO_CODECS,
 	PcmAudioCodec,
 	SubtitleCodec,
@@ -293,7 +293,7 @@ export class MatroskaMuxer extends Muxer {
 					const header = parseOpusIdentificationHeader(bytes);
 
 					// Use the preSkip value from the header
-					seekPreRollNs = Math.round(1e9 * (header.preSkip / OPUS_INTERNAL_SAMPLE_RATE));
+					seekPreRollNs = Math.round(1e9 * (header.preSkip / OPUS_SAMPLE_RATE));
 				}
 			}
 

src/media-sink.ts

@@ -7,6 +7,7 @@
  */
 
 import { parsePcmCodec, PCM_AUDIO_CODECS, PcmAudioCodec, VideoCodec, AudioCodec } from './codec';
+import { extractHevcNalUnits, extractNalUnitTypeForHevc, HevcNalUnitType } from './codec-data';
 import { CustomVideoDecoder, customVideoDecoders, CustomAudioDecoder, customAudioDecoders } from './custom-coder';
 import { InputAudioTrack, InputTrack, InputVideoTrack } from './input-track';
 import {
@@ -624,8 +625,8 @@ export abstract class BaseMediaSampleSink<
 					const nextPacket = await packetSink.getNextPacket(currentPacket);
 					assert(nextPacket);
 
-					currentPacket = nextPacket;
 					decoder.decode(nextPacket);
+					currentPacket = nextPacket;
 				}
 
 				maxSequenceNumber = -1;
@@ -757,12 +758,14 @@ class VideoDecoderWrapper extends DecoderWrapper<VideoSample> {
 
 	inputTimestamps: number[] = []; // Timestamps input into the decoder, sorted.
 	sampleQueue: VideoSample[] = []; // Safari-specific thing, check usage.
+	currentPacketIndex = 0;
+	raslSkipped = false; // For HEVC stuff
 
 	constructor(
 		onSample: (sample: VideoSample) => unknown,
 		onError: (error: DOMException) => unknown,
-		codec: VideoCodec,
+		public codec: VideoCodec,
-		decoderConfig: VideoDecoderConfig,
+		public decoderConfig: VideoDecoderConfig,
 		public rotation: Rotation,
 		public timeResolution: number,
 	) {
@@ -848,6 +851,26 @@ class VideoDecoderWrapper extends DecoderWrapper<VideoSample> {
 	}
 
 	decode(packet: EncodedPacket) {
+		if (this.codec === 'hevc' && this.currentPacketIndex > 0 && !this.raslSkipped) {
+			// If we're using HEVC, we need to make sure to skip any RASL slices that follow a non-IDR key frame such as
+			// CRA_NUT. This is because RASL slices cannot be decoded without data before the CRA_NUT. Browsers behave
+			// differently here: Chromium drops the packets, Safari throws a decoder error. Either way, it's not good
+			// and causes bugs upstream. So, let's take the dropping into our own hands.
+			const nalUnits = extractHevcNalUnits(packet.data, this.decoderConfig);
+			const hasRaslPicture = nalUnits.some((x) => {
+				const type = extractNalUnitTypeForHevc(x);
+				return type === HevcNalUnitType.RASL_N || type === HevcNalUnitType.RASL_R;
+			});
+
+			if (hasRaslPicture) {
+				return; // Drop
+			}
+
+			this.raslSkipped = true;
+		}
+
+		this.currentPacketIndex++;
+
 		if (this.customDecoder) {
 			this.customDecoderQueueSize++;
 			void this.customDecoderCallSerializer
@@ -879,6 +902,9 @@ class VideoDecoderWrapper extends DecoderWrapper<VideoSample> {
 
 			this.sampleQueue.length = 0;
 		}
+
+		this.currentPacketIndex = 0;
+		this.raslSkipped = false;
 	}
 
 	close() {

src/ogg/ogg-demuxer.ts

@@ -6,7 +6,7 @@
  * file, You can obtain one at https://mozilla.org/MPL/2.0/.
  */
 
-import { OPUS_INTERNAL_SAMPLE_RATE } from '../codec';
+import { OPUS_SAMPLE_RATE } from '../codec';
 import { parseModesFromVorbisSetupPacket, parseOpusIdentificationHeader } from '../codec-data';
 import { Demuxer } from '../demuxer';
 import { Input } from '../input';
@@ -249,7 +249,7 @@ export class OggDemuxer extends Demuxer {
 
 		const header = parseOpusIdentificationHeader(firstPacket.data);
 		bitstream.numberOfChannels = header.outputChannelCount;
-		bitstream.sampleRate = header.inputSampleRate;
+		bitstream.sampleRate = OPUS_SAMPLE_RATE; // Always the same
 
 		bitstream.codecInfo.opusInfo = {
 			preSkip: header.preSkip,
@@ -574,7 +574,7 @@ class OggAudioTrackBacking implements InputAudioTrackBacking {
 	constructor(public bitstream: LogicalBitstream, public demuxer: OggDemuxer) {
 		// Opus always uses a fixed sample rate for its internal calculations, even if the actual rate is different
 		this.internalSampleRate = bitstream.codecInfo.codec === 'opus'
-			? OPUS_INTERNAL_SAMPLE_RATE
+			? OPUS_SAMPLE_RATE
 			: bitstream.sampleRate;
 	}
 

src/ogg/ogg-muxer.ts

@@ -6,7 +6,7 @@
  * file, You can obtain one at https://mozilla.org/MPL/2.0/.
  */
 
-import { OPUS_INTERNAL_SAMPLE_RATE, validateAudioChunkMetadata } from '../codec';
+import { OPUS_SAMPLE_RATE, validateAudioChunkMetadata } from '../codec';
 import { parseModesFromVorbisSetupPacket, parseOpusIdentificationHeader } from '../codec-data';
 import {
 	assert,
@@ -119,7 +119,7 @@ export class OggMuxer extends Muxer {
 			track,
 			serialNumber,
 			internalSampleRate: track.source._codec === 'opus'
-				? OPUS_INTERNAL_SAMPLE_RATE
+				? OPUS_SAMPLE_RATE
 				: meta.decoderConfig.sampleRate,
 			codecInfo: {
 				codec: track.source._codec,