fix: Prevent skipping frames when we have garbage data between adts sync words (#390)

Author: brandonocasey

lib/codecs/adts.js

@@ -64,7 +64,7 @@ AdtsStream = function(handlePartialSegments) {
 
     // Prepend any data in the buffer to the input data so that we can parse
     // aac frames the cross a PES packet boundary
-    if (buffer) {
+    if (buffer && buffer.length) {
       oldBuffer = buffer;
       buffer = new Uint8Array(oldBuffer.byteLength + packet.data.byteLength);
       buffer.set(oldBuffer);
@@ -75,8 +75,10 @@ AdtsStream = function(handlePartialSegments) {
 
     // unpack any ADTS frames which have been fully received
     // for details on the ADTS header, see http://wiki.multimedia.cx/index.php?title=ADTS
-    while (i + 5 < buffer.length) {
 
+    // We use i + 7 here because we want to be able to parse the entire header.
+    // If we don't have enough bytes to do that, then we definitely won't have a full frame.
+    while ((i + 7) < buffer.length) {
       // Look for the start of an ADTS header..
       if ((buffer[i] !== 0xFF) || (buffer[i + 1] & 0xF6) !== 0xF0) {
         // If a valid header was not found,  jump one forward and attempt to
@@ -91,6 +93,7 @@ AdtsStream = function(handlePartialSegments) {
 
       // Frame length is a 13 bit integer starting 16 bits from the
       // end of the sync sequence
+      // NOTE: frame length includes the size of the header
       frameLength = ((buffer[i + 3] & 0x03) << 11) |
         (buffer[i + 4] << 3) |
         ((buffer[i + 5] & 0xe0) >> 5);
@@ -99,12 +102,10 @@ AdtsStream = function(handlePartialSegments) {
       adtsFrameDuration = (sampleCount * ONE_SECOND_IN_TS) /
         ADTS_SAMPLING_FREQUENCIES[(buffer[i + 2] & 0x3c) >>> 2];
 
-      frameEnd = i + frameLength;
-
-      // If we don't have enough data to actually finish this ADTS frame, return
-      // and wait for more data
-      if (buffer.byteLength < frameEnd) {
-        return;
+      // If we don't have enough data to actually finish this ADTS frame,
+      // then we have to wait for more data
+      if ((buffer.byteLength - i) < frameLength) {
+        break;
       }
 
       // Otherwise, deliver the complete AAC frame
@@ -119,20 +120,16 @@ AdtsStream = function(handlePartialSegments) {
         samplingfrequencyindex: (buffer[i + 2] & 0x3c) >>> 2,
         // assume ISO/IEC 14496-12 AudioSampleEntry default of 16
         samplesize: 16,
-        data: buffer.subarray(i + 7 + protectionSkipBytes, frameEnd)
+        // data is the frame without it's header
+        data: buffer.subarray(i + 7 + protectionSkipBytes, i + frameLength)
       });
 
       frameNum++;
-
-      // If the buffer is empty, clear it and return
-      if (buffer.byteLength === frameEnd) {
-        buffer = undefined;
-        return;
-      }
-
-      // Remove the finished frame from the buffer and start the process again
-      buffer = buffer.subarray(frameEnd);
+      i += frameLength;
     }
+
+    // remove processed bytes from the buffer.
+    buffer = buffer.subarray(i);
   };
 
   this.flush = function() {

test/transmuxer.test.js

@@ -2729,6 +2729,57 @@ QUnit.test('generates AAC frame events from ADTS bytes', function(assert) {
   assert.equal(frames[0].samplesize, 16, 'parsed samplesize');
 });
 
+QUnit.test('skips garbage data between sync words', function(assert) {
+  var frames = [];
+  adtsStream.on('data', function(frame) {
+    frames.push(frame);
+  });
+
+  var frameHeader = [
+    0xff, 0xf1,       // no CRC
+    0x10,             // AAC Main, 44.1KHz
+    0xbc, 0x01, 0x20, // 2 channels, frame length 9 including header
+    0x00,             // one AAC per ADTS frame
+  ];
+  adtsStream.push({
+    type: 'audio',
+    data: new Uint8Array(
+      []
+      // garbage
+      .concat([0x00, 0x00, 0x00])
+      // frame
+      .concat(frameHeader)
+      .concat([0x00, 0x01])
+      // garbage
+      .concat([0x00, 0x00, 0x00, 0x00, 0x00])
+      .concat(frameHeader)
+      .concat([0x00, 0x02])
+      // garbage
+      .concat([0x00, 0x00, 0x00, 0x00])
+      .concat(frameHeader)
+      .concat([0x00, 0x03])
+      .concat([0x00, 0x00, 0x00, 0x00])
+    )
+  });
+
+  assert.equal(frames.length, 3, 'generated three frames');
+  frames.forEach(function(frame, i) {
+    assert.deepEqual(
+      new Uint8Array(frame.data),
+      new Uint8Array([0x00, i + 1]),
+      'extracted AAC frame'
+    );
+
+    assert.equal(frame.channelcount, 2, 'parsed channelcount');
+    assert.equal(frame.samplerate, 44100, 'parsed samplerate');
+
+    // Chrome only supports 8, 16, and 32 bit sample sizes. Assuming the
+    // default value of 16 in ISO/IEC 14496-12 AudioSampleEntry is
+    // acceptable.
+    assert.equal(frame.samplesize, 16, 'parsed samplesize');
+  });
+});
+
 QUnit.test('parses across packets', function(assert) {
   var frames = [];
   adtsStream.on('data', function(frame) {