Function parseSidx

packages/hifi/src/fmp4/parser.ts:54–100

(
  data: Uint8Array,
  boxOffset: number,
  boxSize: number,
)

// YouTube serves fMP4 audio with a sidx box that maps each fragment
// to a byte range and duration, which we need for MSE seeking.
export function parseSidx(
  data: Uint8Array,
  boxOffset: number,
  boxSize: number,
): { references: SegmentReference[]; timescale: number } {
  const reader = new BinaryReader(data, boxOffset + BOX_HEADER_SIZE);

  const version = reader.readUint8();
  reader.skip(3); // flags

  reader.skip(4); // reference_ID
  const timescale = reader.readUint32();

  let firstOffset: number;
  if (version === 0) {
    reader.skip(4); // earliest_presentation_time (32-bit)
    firstOffset = reader.readUint32();
  } else {
    reader.skip(8); // earliest_presentation_time (64-bit)
    firstOffset = reader.readUint64();
  }

  reader.skip(2); // reserved
  const referenceCount = reader.readUint16();

  let byteOffset = boxOffset + boxSize + firstOffset;
  let timeOffset = 0;
  const references: SegmentReference[] = [];

  for (let index = 0; index < referenceCount; index++) {
    const referencedSize = reader.readUint32() & 0x7fffffff;
    const subsegmentDuration = reader.readUint32();
    reader.skip(4); // starts_with_SAP + SAP_type + SAP_delta_time

    const startByte = byteOffset;
    const endByte = byteOffset + referencedSize - 1;
    const startTime = timeOffset / timescale;
    const endTime = (timeOffset + subsegmentDuration) / timescale;

    references.push({ startByte, endByte, startTime, endTime });

    byteOffset += referencedSize;
    timeOffset += subsegmentDuration;
  }

  return { references, timescale };
}

export function parseInitSegment(data: Uint8Array): Fmp4Index {
  const boxes = findBoxes(data);