/*! @license
* Shaka Player
* Copyright 2016 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
goog.provide('shaka.cea.Mp4CeaParser');
goog.require('goog.asserts');
goog.require('shaka.cea.ICeaParser');
goog.require('shaka.cea.SeiProcessor');
goog.require('shaka.util.DataViewReader');
goog.require('shaka.util.Error');
goog.require('shaka.util.Mp4Parser');
goog.require('shaka.util.Mp4BoxParsers');
/**
* MPEG4 stream parser used for extracting 708 closed captions data.
* @implements {shaka.cea.ICeaParser}
*/
shaka.cea.Mp4CeaParser = class {
/** */
constructor() {
/**
* SEI data processor.
* @private
* @const {!shaka.cea.SeiProcessor}
*/
this.seiProcessor_ = new shaka.cea.SeiProcessor();
/**
* Map of track id to corresponding timescale.
* @private {!Map<number, number>}
*/
this.trackIdToTimescale_ = new Map();
/**
* Default sample duration, as specified by the TREX box.
* @private {!number}
*/
this.defaultSampleDuration_ = 0;
/**
* Default sample size, as specified by the TREX box.
* @private {!number}
*/
this.defaultSampleSize_ = 0;
}
/**
* Parses the init segment. Gets Default Sample Duration and Size from the
* TREX box, and constructs a map of Track IDs to timescales. Each TRAK box
* contains a track header (TKHD) containing track ID, and a media header box
* (MDHD) containing the timescale for the track
* @override
*/
init(initSegment) {
const Mp4Parser = shaka.util.Mp4Parser;
const trackIds = [];
const timescales = [];
new Mp4Parser()
.box('moov', Mp4Parser.children)
.box('mvex', Mp4Parser.children)
.fullBox('trex', (box) => {
const parsedTREXBox = shaka.util.Mp4BoxParsers.parseTREX(
box.reader);
this.defaultSampleDuration_ = parsedTREXBox.defaultSampleDuration;
this.defaultSampleSize_ = parsedTREXBox.defaultSampleSize;
})
.box('trak', Mp4Parser.children)
.fullBox('tkhd', (box) => {
goog.asserts.assert(
box.version != null,
'TKHD is a full box and should have a valid version.');
const parsedTKHDBox = shaka.util.Mp4BoxParsers.parseTKHD(
box.reader, box.version);
trackIds.push(parsedTKHDBox.trackId);
})
.box('mdia', Mp4Parser.children)
.fullBox('mdhd', (box) => {
goog.asserts.assert(
box.version != null,
'MDHD is a full box and should have a valid version.');
const parsedMDHDBox = shaka.util.Mp4BoxParsers.parseMDHD(
box.reader, box.version);
timescales.push(parsedMDHDBox.timescale);
})
.parse(initSegment, /* partialOkay= */ true);
// At least one track should exist, and each track should have a
// corresponding Id in TKHD box, and timescale in its MDHD box
if (!trackIds.length|| !timescales.length ||
trackIds.length != timescales.length) {
throw new shaka.util.Error(
shaka.util.Error.Severity.CRITICAL,
shaka.util.Error.Category.TEXT,
shaka.util.Error.Code.INVALID_MP4_CEA);
}
// Populate the map from track Id to timescale
trackIds.forEach((trackId, idx) => {
this.trackIdToTimescale_.set(trackId, timescales[idx]);
});
}
/**
* Parses each video segment. In fragmented MP4s, MOOF and MDAT come in
* pairs. The following logic gets the necessary info from MOOFs to parse
* MDATs (base media decode time, sample sizes/offsets/durations, etc),
* and then parses the MDAT boxes for CEA-708 packets using this information.
* CEA-708 packets are returned in the callback.
* @override
*/
parse(mediaSegment) {
const Mp4Parser = shaka.util.Mp4Parser;
/** @type {!Array<!shaka.cea.ICeaParser.CaptionPacket>} **/
const captionPackets = [];
// Fields that are found in MOOF boxes
let defaultSampleDuration = this.defaultSampleDuration_;
let defaultSampleSize = this.defaultSampleSize_;
let sampleData = [];
let baseMediaDecodeTime = null;
let timescale = shaka.cea.ICeaParser.DEFAULT_TIMESCALE_VALUE;
new Mp4Parser()
.box('moof', Mp4Parser.children)
.box('traf', Mp4Parser.children)
.fullBox('trun', (box) => {
goog.asserts.assert(
box.version != null && box.flags!=null,
'TRUN is a full box and should have a valid version & flags.');
const parsedTRUN = shaka.util.Mp4BoxParsers.parseTRUN(
box.reader, box.version, box.flags);
sampleData = parsedTRUN.sampleData;
})
.fullBox('tfhd', (box) => {
goog.asserts.assert(
box.flags != null,
'TFHD is a full box and should have valid flags.');
const parsedTFHD = shaka.util.Mp4BoxParsers.parseTFHD(
box.reader, box.flags);
// If specified, defaultSampleDuration and defaultSampleSize
// override the ones specified in the TREX box
defaultSampleDuration = parsedTFHD.defaultSampleDuration ||
this.defaultSampleDuration_;
defaultSampleSize = parsedTFHD.defaultSampleSize ||
this.defaultSampleSize_;
const trackId = parsedTFHD.trackId;
// Get the timescale from the track Id
if (this.trackIdToTimescale_.has(trackId)) {
timescale = this.trackIdToTimescale_.get(trackId);
}
})
.fullBox('tfdt', (box) => {
goog.asserts.assert(
box.version != null,
'TFDT is a full box and should have a valid version.');
const parsedTFDT = shaka.util.Mp4BoxParsers.parseTFDT(
box.reader, box.version);
baseMediaDecodeTime = parsedTFDT.baseMediaDecodeTime;
})
.box('mdat', (box) => {
if (baseMediaDecodeTime === null) {
// This field should have been populated by
// the Base Media Decode time in the TFDT box
throw new shaka.util.Error(
shaka.util.Error.Severity.CRITICAL,
shaka.util.Error.Category.TEXT,
shaka.util.Error.Code.INVALID_MP4_CEA);
}
this.parseMdat_(box.reader, baseMediaDecodeTime, timescale,
defaultSampleDuration, defaultSampleSize, sampleData,
captionPackets);
})
.parse(mediaSegment, /* partialOkay= */ false);
return captionPackets;
}
/**
* Parse MDAT box.
* @param {!shaka.util.DataViewReader} reader
* @param {!number} time
* @param {!number} timescale
* @param {!number} defaultSampleDuration
* @param {!number} defaultSampleSize
* @param {!Array<shaka.util.ParsedTRUNSample>} sampleData
* @param {!Array<!shaka.cea.ICeaParser.CaptionPacket>} captionPackets
* @private
*/
parseMdat_(reader, time, timescale, defaultSampleDuration,
defaultSampleSize, sampleData, captionPackets) {
let sampleIndex = 0;
// The fields in each ParsedTRUNSample contained in the sampleData
// array are nullable. In the case of sample data and sample duration,
// we use the defaults provided by the TREX/TFHD boxes. For sample
// composition time offset, we default to 0.
let sampleSize = defaultSampleSize;
if (sampleData.length) {
sampleSize = sampleData[0].sampleSize || defaultSampleSize;
}
while (reader.hasMoreData()) {
const naluSize = reader.readUint32();
const naluType = reader.readUint8() & 0x1F;
if (naluType == shaka.cea.ICeaParser.NALU_TYPE_SEI) {
let timeOffset = 0;
if (sampleData.length > sampleIndex) {
timeOffset = sampleData[sampleIndex].sampleCompositionTimeOffset || 0;
}
const pts = (time + timeOffset)/timescale;
for (const packet of this.seiProcessor_
.process(reader.readBytes(naluSize - 1))) {
captionPackets.push({
packet,
pts,
});
}
} else {
try {
reader.skip(naluSize - 1);
} catch (e) {
// It is necessary to ignore this error because it can break the start
// of playback even if the user does not want to see the subtitles.
break;
}
}
sampleSize -= (naluSize + 4);
if (sampleSize == 0) {
if (sampleData.length > sampleIndex) {
time += sampleData[sampleIndex].sampleDuration ||
defaultSampleDuration;
} else {
time += defaultSampleDuration;
}
sampleIndex++;
if (sampleData.length > sampleIndex) {
sampleSize = sampleData[sampleIndex].sampleSize || defaultSampleSize;
} else {
sampleSize = defaultSampleSize;
}
}
}
}
};