From 022b157eeb941f7028884fcd1add9ceecbc5d138 Mon Sep 17 00:00:00 2001
From: Vanilagy <1696106+Vanilagy@users.noreply.github.com>
Date: Thu, 2 Jan 2025 19:28:43 +0100
Subject: [PATCH] Turn AudioBuffers into AudioData in a chunked fashion

---
 .gitignore                     |  3 +-
 dev/index.html                 |  4 ++-
 src/isobmff/isobmff-demuxer.ts |  2 +-
 src/media-source.ts            | 59 ++++++++++++++++++++++------------
 todo.txt                       | 10 ++++++
 5 files changed, 53 insertions(+), 25 deletions(-)
 create mode 100644 todo.txt

diff --git a/.gitignore b/.gitignore
index d090bb7..57d3105 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,4 +1,3 @@
 /node_modules
 /build
-/dist
-todo.txt
\ No newline at end of file
+/dist
\ No newline at end of file
diff --git a/dev/index.html b/dev/index.html
index f4aeb59..69808ee 100644
--- a/dev/index.html
+++ b/dev/index.html
@@ -40,6 +40,7 @@
 	format = new Metamuxer.Mp4OutputFormat({ fastStart: false }); // new Metamuxer.MkvOutputFormat();// new Metamuxer.Mp4OutputFormat({ fastStart: false });
 	let target = new Metamuxer.ArrayBufferTarget();
 
+	/*
 	target = new Metamuxer.StreamTarget(new WritableStream({
 		async write(chunk) {
 			console.log(chunk)
@@ -47,6 +48,7 @@
 			await new Promise(resolve => setTimeout(resolve, 250));
 		}
 	}), { chunked: true });
+	*/
 
 	let output = new Metamuxer.Output({ format, target });
 
@@ -170,5 +172,5 @@ Testing... <00:17.350>One... <00:18.125>Two...
 	await output.finalize();
 
 	console.log(target);
-	//download(new Blob([target.buffer]), 'test.mp4');
+	download(new Blob([target.buffer]), 'test.mp4');
 </script>
\ No newline at end of file
diff --git a/src/isobmff/isobmff-demuxer.ts b/src/isobmff/isobmff-demuxer.ts
index 640e5a6..3a781b1 100644
--- a/src/isobmff/isobmff-demuxer.ts
+++ b/src/isobmff/isobmff-demuxer.ts
@@ -1076,7 +1076,7 @@ export class IsobmffDemuxer extends Demuxer {
 
 				const startPos = this.isobmffReader.pos;
 
-				while (true) {
+				while (this.isobmffReader.pos < boxEndPos) {
 					const flagAndType = this.isobmffReader.readU8();
 					const metadataBlockLength = this.isobmffReader.readU24();
 					const type = flagAndType & BLOCK_TYPE_MASK;
diff --git a/src/media-source.ts b/src/media-source.ts
index a6d34d5..ae72bf8 100644
--- a/src/media-source.ts
+++ b/src/media-source.ts
@@ -517,32 +517,49 @@ export class AudioBufferSource extends AudioSource {
 			throw new TypeError('audioBuffer must be an AudioBuffer.');
 		}
 
+		const MAX_FLOAT_COUNT = 64 * 1024 * 1024;
+
 		const numberOfChannels = audioBuffer.numberOfChannels;
 		const sampleRate = audioBuffer.sampleRate;
-		const numberOfFrames = audioBuffer.length;
+		const totalFrames = audioBuffer.length;
+		const maxFramesPerChunk = Math.floor(MAX_FLOAT_COUNT / numberOfChannels);
 
-		// Create a planar F32 array containing all channels
-		const data = new Float32Array(numberOfChannels * numberOfFrames);
-		for (let channel = 0; channel < numberOfChannels; channel++) {
-			const channelData = audioBuffer.getChannelData(channel);
-			data.set(channelData, channel * numberOfFrames);
+		let currentRelativeFrame = 0;
+		let remainingFrames = totalFrames;
+
+		const promises: Promise<void>[] = [];
+
+		// Create AudioData in a chunked fashion so we don't create huge Float32Arrays
+		while (remainingFrames > 0) {
+			const framesToCopy = Math.min(maxFramesPerChunk, remainingFrames);
+			const chunkData = new Float32Array(numberOfChannels * framesToCopy);
+
+			for (let channel = 0; channel < numberOfChannels; channel++) {
+				audioBuffer.copyFromChannel(
+					chunkData.subarray(channel * framesToCopy, channel * framesToCopy + framesToCopy),
+					channel,
+					currentRelativeFrame,
+				);
+			}
+
+			const audioData = new AudioData({
+				format: 'f32-planar',
+				sampleRate,
+				numberOfFrames: framesToCopy,
+				numberOfChannels,
+				timestamp: (1e6 * (this._accumulatedFrameCount + currentRelativeFrame)) / sampleRate,
+				data: chunkData,
+			});
+
+			promises.push(this._encoder.digest(audioData));
+			audioData.close();
+
+			currentRelativeFrame += framesToCopy;
+			remainingFrames -= framesToCopy;
 		}
 
-		const audioData = new AudioData({
-			format: 'f32-planar',
-			sampleRate,
-			numberOfFrames,
-			numberOfChannels,
-			timestamp: Math.round(1e6 * this._accumulatedFrameCount / sampleRate),
-			data: data,
-		});
-
-		const promise = this._encoder.digest(audioData);
-		audioData.close();
-
-		this._accumulatedFrameCount += numberOfFrames;
-
-		return promise;
+		this._accumulatedFrameCount += totalFrames;
+		return Promise.all(promises);
 	}
 
 	/** @internal */
diff --git a/todo.txt b/todo.txt
new file mode 100644
index 0000000..54f50c9
--- /dev/null
+++ b/todo.txt
@@ -0,0 +1,10 @@
+- Manual key frame passing
+- Progress reporting
+- Test tree shaking
+- Throw if edit list detected
+- Audio codec string regex validation
+- Add the new audio codecs to muxers
+- Mov muxer!!! Only mov can hold PCM audio, MP4 cannot
+- Metadata methods for computing average fps and bitrate
+- close method on Output to cancel it (and dispose the encoders). Perhaps abort is a better title
+- lower default key frame interval (3 secs is better I think?) and make it outside-configurable
\ No newline at end of file