mediabunny 1.66.0 on Node.js NPM

This release is the result of #636 and includes over 400 improvements and bug fixes spanning almost every area of Mediabunny. The goal was to make Mediabunny maximally robust and to fix its bugs before anyone runs into them. Many of these flaws were found by a thorough LLM-assited audit of the entire codebase which took multiple weeks to fully review and test. The result is a library that's more reliable, resilient, correct, and should "just work".

InputTrack.id is now the track's 1-based index in getTracks() for every format, making it unique in multi-segment Matroska files. Previously, it was the container's own track ID (such as the MPEG-TS PID or the Ogg serial number).
Added InputTrack.getFirstTimestampOrNull() and InputTrack.computeDurationOrNull(), which return null instead of 0 for tracks without packets
Added WavOutputFormatOptions.appendOnly for writing WAVE files in an append-only fashion
Added support for setting HlsOutputFormatOptions.targetDuration to Infinity outside of live mode, giving each playlist only a single segment
Added support for hybrid fragmented MP4 files, which hold samples both in the moov box and in subsequent fragments (such as FFmpeg's -movflags frag_keyframe output). Previously, all fragments were ignored.
Added read support for 64-bit float PCM in ISOBMFF
Added read support for MP3 in MOV files with a .mp3 sample entry
Added support for ISOBMFF files without an ftyp box
Added support for the ISOBMFF srat box, fixing high sample rates being misreported for ipcm/fpcm tracks
Added support for reading the Matroska segment title (Info\Title) as metadata
Added support for E-AC-3 dependent substreams in MPEG-TS, fixing 7.1 E-AC-3 being read as 6 channels with doubled timing
Added support for reading Ogg start trimming
Added support for VBRI headers in MP3 files
Added support for AAC with a program config element (channel configuration 0) in ADTS and MPEG-TS
Added support for writing 6.1 and 7.1 AAC to ADTS and MPEG-TS
Added detection of ADTS and MP3 files holding just a single frame
Added support for ADTS files with junk before or between frames
Improved input format detection
Improved tree shaking
Added support for ID3v2.2 TXX frames
Added support for fLaC in HLS CODECS attributes
Fixed MOV output writing sample rates of 65536 Hz and above as 0
Fixed reading sample rates of 65536 Hz and above from ISOBMFF files
Fixed ISOBMFF PCM tracks losing most of their data when their samples span more than one PCM frame
Fixed ISOBMFF fragments without tfdt drifting by the composition delay with every fragment
Fixed concurrent reads of the same ISOBMFF fragment returning a partially processed fragment, which could surface undecrypted data
Fixed all-ones mdhd durations not being treated as unknown
Fixed saio offsets in fragments being resolved relative to the moof box instead of the base data offset
Fixed wrong sample data in fragmented MP4 files without base data offsets when an unsupported track comes before
Fixed pssh boxes differing only in their key IDs being deduplicated
Fixed fragmented MP4 muxer marking key frames in the middle of a fragment as non-key frames
Fixed fragmented MP4 muxer sometimes writing the moov box before all tracks' start offsets were known, leading to missing edit lists and A/V desync
Fixed fragmented MP4 mdat boxes of 4 GiB and larger being written with the wrong size
Fixed rare corrupt chunk offsets in MP4 files over 4 GiB written with fastStart: 'in-memory'
ISOBMFF muxer now throws for fragments whose track data begins 2 GiB or more after the moof box, and for CMAF segments of 2 GiB or more or longer than about 20.7 hours, instead of silently writing broken offsets and durations
Fixed onMoof reporting incorrectly shifted fragment start timestamps
Fixed fastStart: 'in-memory' not working with AppendOnlyStreamTarget
Fixed fastStart: 'reserve' including the trailing free box in the data passed to onMoov
Reduced peak memory use when finalizing MP4 files with fastStart: 'in-memory' into a StreamTarget
Fixed WebVTT cues with zero duration hanging the ISOBMFF muxer
Fixed touching WebVTT cues producing zero-length samples in ISOBMFF
Fixed the WebVTT ctim box being written in seconds instead of milliseconds
Fixed fragmented MP4 output failing to finalize with a short WebVTT or E-AC-3 track
Fixed disc number being written under the disc key instead of disk
Fixed PCM with more than 16 bits per sample in MP4 being written with a QuickTime-only sound description, making it unplayable in AVFoundation
AC-3 and E-AC-3 in MOV are now wrapped in a wave box, fixing E-AC-3 playback in AVFoundation
MP3 in MOV is now written with a .mp3 sample entry, fixing playback in AVFoundation
Fixed the ISOBMFF muxer hanging on finalize() after start() or a track close threw
Fixed VP8's vpcC box being written with a bit depth of 0
Fixed MOV hdlr component names being written as C strings instead of counted strings
Fixed the sample entry of FLAC in MP4 always stating 16 bits per sample (making browsers reject 24-bit FLAC) and a sample rate of 0 above 65535 Hz (#630)
Fixed HE-AAC in Matroska reporting half its sample rate, as OutputSamplingFrequency was ignored
Fixed reading AAC from old Matroska files with legacy codec IDs like A_AAC/MPEG4/LC/SBR and no CodecPrivate
Fixed TrueHD in Matroska taking the container's key flags at face value, which could make seeking start without a restart point
Fixed Matroska audio in blocks with references being read as delta packets
Matroska muxer now writes AAC's core sample rate as SamplingFrequency and adds OutputSamplingFrequency when SBR doubles it
Matroska AV1 codec info is now read from the av1C in CodecPrivate, falling back to the first packet only when it holds no sequence header
Matroska demuxer now respects DisplayUnit, and a lone DisplayWidth or DisplayHeight is no longer dropped
Fixed wrong display dimensions for stereoscopic Matroska video
Fixed a Matroska cluster cut off by the end of the file being dropped entirely
Fixed wrong track numbers in multi-segment Matroska files
Fixed an unknown-size Matroska segment followed by another file (such as concatenated live recordings) reading that file's clusters
Fixed Matroska projection yaw, pitch and roll being applied as 2D flips and rotation to 360-degree video
Fixed the Matroska projection pose being misread when its elements come before ProjectionType
Fixed laced Matroska frames getting timestamps off the tick grid, which could make getPacket(packet.timestamp) return the previous packet
Fixed Matroska files with a DocTypeVersion above 4 (such as mkvmerge's v5 output) being rejected
Fixed Matroska packet durations derived from DefaultDuration not being whole ticks
Fixed empty Matroska elements not taking their default values (an empty TimestampScale threw, an empty FlagDefault read as 0)
Fixed an empty Matroska FileUID throwing
Fixed EBML lacing with a single frame throwing
Fixed EBML unknown sizes of 2 to 7 bytes not being recognized
Fixed Matroska muxer writing a negative segment duration when all timestamps are negative
Fixed Matroska muxer writing corrupt files when an element such as an attachment or a frame was 256 MiB or larger
Fixed Matroska muxer writing BlockDuration on the last block in decode order instead of presentation order, giving the last frame a duration of 0 with B-frames
Matroska muxer now writes MaxBlockAdditionID for tracks with block additions (video with alpha, WebVTT)
WebVTT in WebM output is now written the way the WebM spec defines it (D_WEBVTT/SUBTITLES)
Fixed the Matroska muxer hanging on finalize() after start(), a track close or finalize() threw
Fixed the validation error for MkvOutputFormatOptions.onSegmentHeader naming the wrong option
Matroska muxer no longer patches the color space into VP9 key frames, which zeroed the color space for BT.2020, corrupted RGB streams and modified the caller's packet data
Fixed MPEG-TS muxer writing a DTS greater than the PTS for B-frames, which made FFmpeg reject the files
Fixed MPEG-TS muxer writing non-increasing DTS when B-frames first appear later in a video
Fixed malformed HEVC access unit delimiters in MPEG-TS output, which broke FFmpeg's bitstream filters
Fixed HEVC access unit delimiters in MPEG-TS output having the wrong temporal layer for temporally scalable video
Fixed MPEG-TS PIDs depending on which track's data came first, which made them change between HLS segments
Fixed a rejected first AAC packet leaving the MPEG-TS track half set up, making every following packet fail
Fixed the MPEG-TS muxer hanging on finalize() after start() or a track close threw
Fixed PAT and PMT sections with a pointer field losing the program or its last stream
Fixed private sections on the PMT PID being parsed as the PMT
Fixed the PMT still being guessed from other PIDs after the PAT had named it
Fixed MPEG-TS captures joined mid-broadcast truncating a PES packet that straddles the PAT and PMT
Fixed MPEG-TS packet lookups near the start of a track reading data from before the PMT
Fixed MPEG-TS init inputs not being read before use, which made segments without parameter sets throw
Fixed a race in MPEG-TS getNextPacket that could skip a packet when called concurrently for the same packet (#640)
Fixed wrong MPEG-TS packet durations with B-frames when retrieving the next packet of the same packet a second time
Fixed getPacket returning null in MPEG-TS for open-GOP leading frames, which are presented before the first packet in decode order
Fixed a multi-packet non-PES section on a track's PID ending the MPEG-TS stream early
Fixed false ADTS sync words in AAC payloads ending MPEG-TS audio streams early
Fixed a tiny final AVC slice in MPEG-TS being dropped or merged into the previous frame
Fixed HEVC end of sequence and end of bitstream NAL units starting a new frame in MPEG-TS
Fixed audio timestamps drifting within MPEG-TS PES packets that hold several frames
Fixed MPEG-TS packets without a successor ignoring the codec's frame duration (a lone AAC packet had a duration of 0)
Fixed the last key packet of intra-only AVC in MPEG-TS having a duration of 0
Fixed negative MPEG-TS timestamps following a positive first timestamp being read as about 95443 seconds
Fixed MPEG-TS audio frames only becoming available once the next PES packet began, delaying live streams (#635)
Ogg muxer now writes streams that begin before the codec's usual start instead of shifting them forward. Packets too far before the start to be represented are dropped with a warning.
OggOutputFormat.negativeTimestampSupport is now 'prefer-non-negative', allowing forced copy conversions with trims into Ogg
Removed end trimming from Ogg reading and writing, as packet durations can't properly signal it
Ogg demuxer now throws for empty packets in the middle of a stream, which no supported codec allows (the empty final packet VLC writes is still supported)
Fixed the Vorbis comment header in Ogg output lacking its framing bit, making libvorbis reject the files
Fixed the Ogg muxer interleaving tracks with different sample rates by sample counts instead of time
Fixed the Ogg muxer writing header pages after data pages in multi-track files
Fixed the last page of an Ogg track that ended early being held back until finalization
Fixed the Ogg muxer hanging on finalize() after start() or a track close threw
Fixed Ogg seeking using the granule position of the page a packet starts on instead of the one it ends on
Fixed Ogg packet lookups returning the wrong packet in files with a broken page chain
Fixed seeking to the last Vorbis packet ahead of an empty final packet (as written by VLC) giving it the wrong duration
Fixed truncated Ogg files reporting a duration of 0 and failing to find their last packet
Fixed the Ogg MIME type reading codecs="" for streams without known codecs
FLAC muxer now supports total sample counts of 2^32 and above
FLAC muxer now throws for PICTURE and Vorbis comment blocks larger than 16 MiB instead of writing broken block lengths
Fixed FLAC files at 12 or 64 kHz having no packets
Fixed FLAC total sample counts of 2^32 and above being misread
Fixed FLAC frame sample numbers of 2^31 and above being misread
Fixed concurrent getFirstPacket() calls on FLAC files corrupting the packet index
Fixed FLAC getPacket() sometimes returning the previous packet due to floating point error
Fixed MP3, ADTS and FLAC getPacket() at a packet's end timestamp sometimes returning that packet instead of the next one
Fixed broken FLAC PICTURE blocks for MIME types or descriptions with non-ASCII characters
Fixed single-frame FLAC output having a minimum block size of 0, which FFmpeg rejects
Fixed missing validation for FlacOutputFormatOptions.onFrame
Fixed reading WAVE files whose header overstates the data size from unsized sources, such as FFmpeg's pipe output
Fixed WAVE files with a placeholder data size reporting a huge duration from metadata
Fixed unfinalized WAVE files, whose sizes are still 0, failing to load
Fixed WAVE files over 4 GiB with placeholder sizes (as written by FFmpeg) being cut off at 4 GiB
Fixed chunks following the data chunk in RF64 files, such as metadata, not being read
Fixed WAVE getPacket() returning the previous packet when given a packet's exact timestamp
Fixed RIFF INFO years being read in local time instead of UTC
Fixed UTF-8 RIFF INFO strings, such as the ones FFmpeg writes, being misread
Fixed the WAVE muxer not padding odd-sized data chunks
WAVE muxer now writes WAVE_FORMAT_EXTENSIBLE with a channel mask for more than two channels and for integer PCM above 16 bits
WAVE muxer now writes a fact chunk for float, A-law and mu-law audio, except in append-only mode
Fixed the Xing/Info frame count of MP3 output including the Xing frame itself, making every file one frame too long
Fixed the Xing frame of MP3 output claiming a CRC it didn't have
The Xing frame of MP3 output now has the same bitrate as the first audio frame where possible, so that CBR files are constant bitrate throughout
MP3 muxer no longer writes a Xing frame for Layer I and II audio, where decoders would play it as audio
Fixed MP3 frame headers with a reserved MPEG version or sample rate index being accepted, causing false syncs
Fixed MP3 files starting with junk that resembles a frame header not being detected
Fixed junk between MP3 frames being read as frames
Fixed trailing ID3v1 and APEv2 tags counting toward the estimated duration of CBR MP3 files
ADTS muxer now throws for AAC configurations and frames that ADTS can't represent instead of writing corrupt headers
Fixed ADTS reading stopping at ID3v2 tags between frames
ADTS and MP3 streams now end with a warning where the sample rate changes, since a track can only have one sample rate
ID3v2 COMM frames with a description (like iTunNORM) are no longer read as the comment
Fixed ID3v2.2 and ID3v2.3 tags with whole-tag unsynchronisation being misread (#631)
Fixed ID3v2.4 tags with non-synchsafe frame sizes (as written by iTunes) being misread
Fixed ID3v2 frames with a grouping identity or data length indicator being misread, which could lose all tags (#631)
Fixed ID3v2.3 tags with the footer flag set being misread
Fixed ID3v2.4 extended header sizes being misread
Fixed compressed ID3v2.2 tags being misread; they're now skipped with a warning
Fixed empty UTF-16 strings in ID3v2 being misread
ID3v2 comment and lyrics frames with an invalid text encoding now throw like other text frames, instead of silently reading as nothing
Fixed ID3v2 cover art MIME types being decoded with the wrong text encoding (#631)
Fixed ID3v2.3 dates being misread, as the year (TYER) and day and month (TDAT) weren't combined
Fixed ID3v2.4 dates with a time being read in local time instead of UTC
Fixed ID3 years below 100 being misread or giving invalid dates
Date tags in Vorbis comments, MP4, Matroska and RIFF INFO are now consistently parsed as ISO 8601
RIFF IDIT tags are no longer read as the date
Fixed invalid ID3 years giving invalid dates
Fixed an ID3 BPM of 0 making conversions fail
Fixed ID3v2 lyrics and cover art with non-Latin-1 text being written with the wrong frame size
Fixed cover art being duplicated when converting MP3 to MP3
Fixed missing validation for AttachedImage.name and AttachedImage.description
Fixed HLS variant streams referencing undefined media groups losing their tracks
Fixed the HLS CHANNELS attribute of object-based audio (such as Dolby Atmos) being read as the channel count
Fixed HLS master playlists with only I-frame variants failing to load
Fixed directly loaded HLS I-frame playlists not reporting their tracks as having only key packets
Fixed several cases of relative URLs in HLS playlists being resolved incorrectly, such as protocol-relative URLs, data: URIs and URLs with query strings
Fixed relative HLS paths starting with ../ losing that part when resolved
Improved HLS query parameter parsing
Fixed sparse fMP4 HLS segments with large timestamps being shifted by a timestamp wrap
Fixed HLS packet lookups before the first segment's playlist time returning null
Fixed HLS packet lookups at a packet's exact timestamp sometimes returning the previous packet
Fixed EXT-X-MAP tags discarding the duration or byte range of the following segment
Fixed live HLS playlists getting a wrong timeline when the window slid past all known segments between playlist reloads
Fixed new segments of live HLS playlists losing their init data after a playlist reload
Fixed wrong timing after a discontinuity in live HLS segments that slid out of the window between playlist reloads
Fixed live HLS playlists jumping to Unix timestamps when date-times first appear after a playlist reload
Fixed ended HLS playlists without segments waiting a target duration on every query
Fixed getUnixTimeForTimestamp() waiting for live HLS playlists to reach the given timestamp
VOD HLS media playlists now all state the longest segment duration as their target duration. Previously, it differed between playlists and never went below targetDuration.
Renditions with the same name in an HLS group now get unique names, like "Audio (2)"
Fixed HLS output with fastStart: 'reserve' MP4 segments failing
Fixed HLS output failing for paired tracks that no single segment format can hold together; they're now split into separate renditions
Fixed HLS I-frame tracks being grouped as video renditions
Fixed HLS MP3 segments each starting with a Xing frame, which players decoded as an extra frame of silence
HLS I-frame playlists now hold one I-frame per segment, as the spec requires
HLS variants now list the codecs of every rendition in their groups in CODECS, as the spec requires
Fixed HLS playlists failing to load when a variant lists several codecs of the same type, such as for a group of renditions with different codecs
Fixed doubled BANDWIDTH values and duplicated CODECS entries for HLS variants whose own playlist is one of their renditions
Fixed HLS BANDWIDTH values being computed over the wrong window for non-integer target durations
Fixed the HLS RESOLUTION attribute using the display aspect ratio as the size (giving 16x9)
Fixed HLS master playlists using WebCodecs codec strings for FLAC and MP3 instead of fLaC and mp4a.40.34
Fixed FORCED=YES being written for HLS audio and video renditions
Fixed floating point error sometimes making the HLS target duration one too high
Fixed the HLS muxer throwing when all tracks of a playlist are closed at once (#633)
Fixed HlsOutputFormatOptions accepting NaN for targetDuration and maxLiveSegmentCount
HLS output format no longer lists WebVTT as supported, as it can't write subtitles yet
Fixed HLS errors for unsupported audio codecs with delta packets (such as TrueHD) suggesting different segment formats, which can't help
Fixed Input.computeDuration() counting tracks without packets as ending at 0
Fixed InputVideoTrack.hasHighDynamicRange() reporting wide-gamut SDR video as HDR
InputTrack.computePacketStats() now reports infinite rates for packets that all sit at a single instant, and reads past targetPacketCount until the packets span some time
Fixed conversions to MP4 with fastStart: 'reserve' throwing for tracks whose packets all sit at a single instant
Fixed frame rate detection missing constant frame rates when the container's timescale can't represent the frame times exactly (such as 60 FPS in Matroska)
Fixed MOV files with a 64-bit ftyp box size being detected as MP4
Fixed a disposed Input throwing an internal error instead of InputDisposedError
Fixed failed source reads being cached, so that later reads never retried
Fixed faulty LRU source cache eviction logic
RangedSource now always reports its specified length as its size, and throws when the underlying data is too short for it
Fixed several cases of reading near the end of a file hanging or failing
Fixed FilePathSource, RangedSource and CustomPathedSource not emitting read events
Fixed UrlSource with a Range header emitting read events at the wrong positions
Fixed FilePathSource leaking file handles in several cases
Fixed FilePathSource not working properly with short reads
Fixed FilePathSource throwing an internal error instead of InputDisposedError when disposed while opening
Fixed CustomSource calling getSize once for every concurrent first read
Fixed CustomSource not cancelling streams returned by read that hold more data than requested
Fixed ReadableStreamSource not cancelling its stream when disposed before reading
Fixed UrlSource ignoring the headers of a Request passed as its URL
Fixed UrlSource requests for other files (like HLS segments) incorrectly inheriting the settings of the root request
Fixed UrlSource with a relative URL throwing outside of a window when the server doesn't support range requests
Fixed UrlSource failing to read empty resources
Fixed UrlSource taking capped range responses for the end of resources of unknown size
Fixed UrlSource retrying failed response reads forever (#638)
Fixed UrlSource keeping requests running after errors
Fixed UrlSource error messages sometimes showing [object Request] instead of the URL
Canceling an output now also closes the streams of StreamTargets that were never started, which leaked file handles for FilePathTarget
Fixed Output.cancel() failing and leaving targets open when a source or target stream had errored
Fixed Output.cancel() during finalization resolving before the resources were released
A failed Output.start() now cancels the output and releases its resources, like a failed finalize() does
Fixed targets requested during cancellation being written to after being closed, or never being closed
Fixed Output.getMimeType() never settling after the output was canceled
Fixed unhandled promise rejections from Output.target with a failing async PathedTarget callback and from closing discarded targets
Fixed FilePathTarget leaking its file handle after a write failure
Fixed a stack overflow when a chunked StreamTarget with a small chunkSize received a large write
Fixed AppendOnlyStreamTarget never emitting finalized
Fixed format callbacks such as onMoof sometimes receiving wrong data
Fixed missing validation for the result of OutputOptions.initTarget
Fixed Quality accepting names of inherited object properties, like 'toString', as quality levels
Quality now scales Opus and Vorbis bitrates with the channel count, fixing multichannel Vorbis failing to encode
Fixed a typo in the error for too many tracks
MediaStreamAudioTrackSource and MediaStreamVideoTrackSource now count queued samples toward the encoder's load, so they drop samples sooner under backpressure instead of queuing them up without limit
Invalid adds to VideoSampleSource, CanvasSource, AudioSampleSource and AudioBufferSource now throw synchronously, like for the packet sources
Fixed media sources accepting data added while they were still closing
Fixed rare race conditions related to closing tracks
Fixed some media source race conditions
Fixed MediaStreamVideoTrackSource failing for frames arriving while it shuts down; they're now dropped, like for audio
Fixed MediaSource.close() failures becoming unhandled rejections and making finalize() hang
Fixed custom encoders never being closed after a failed call
Fixed encoders being created while canceling or closing and never being closed
Fixed muxer errors for packets emitted during the final encoder flush being lost
Fixed Output.finalize() succeeding without a track whose encoder failed to initialize
Fixed errors thrown while handling encoder output being swallowed
Fixed adding audio to an encoder that had failed throwing a misleading error
Fixed the first video frame sometimes not being encoded as a key frame when its timestamp is negative
Fixed sizeChangeBehavior set to 'fill', 'contain' or 'cover' not applying to frames of the original size
Fixed VideoTransformOptions.alpha being ignored
Fixed missing validation for VideoTransformOptions.alpha
Fixed encoding video with alpha losing the source's color space
Fixed CanvasSource and MediaStreamVideoTrackSource leaking a video frame per frame when transform.frameRate is set
Fixed an audio process function returning its input sample closing the caller's sample
Fixed empty audio samples making the encoder throw
Fixed a MediaStreamVideoTrackSource paused before its first frame throwing repeatedly
Fixed MediaStreamVideoTrackSource accepting non-finite frame rates, which could hang
Fixed MediaStreamVideoTrackSource failing to start in Safari
Fixed the audio context fallback of MediaStreamAudioTrackSource never being closed
Fixed VP8 and VP9 video with hidden frames stored as separate packets getting increasingly wrong timestamps (#642)
Fixed video frame timestamps sometimes being shifted after seeking
Fixed the first GOP of AVC and HEVC video with alpha decoding without alpha in Chromium
Fixed alpha merging ignoring backpressure
Fixed video with alpha losing its color space and display size
Fixed decoders sometimes not being closed after errors
Fixed sample iterators sometimes leaving their decoder open after return() or throw(), or when the mapping in canvases() or buffers() threw
Fixed samples queued behind a failed alpha merge never being closed
Fixed PCM sample timestamps drifting, which could duplicate samples
Fixed PCM decoding sometimes returning no samples after a seek
Fixed tiny PCM packets producing empty samples
Fixed AudioSamples created from raw data referencing that data instead of copying it
Fixed VideoSample losing the last chroma column of odd-width subsampled formats when copied
Fixed VideoSamples created from raw data ignoring the offset of their visible rect
Fixed VideoSample.copyTo() with an RGB/BGR conversion corrupting pixels for layouts whose offset or stride isn't a multiple of 4, and overwriting the bytes before the layout's offset
Fixed VideoSample.drawWithFit() not clamping the crop, and leaving the context transformed when throwing
Fixed VideoSample.transform() dropping encode options
Fixed resource leaks when constructing a VideoSample or AudioSample failed, and when converting a VideoSample from YUV to RGB failed
Fixed VideoSample throwing internal errors instead of a TypeError for wrong layout plane counts, empty visible rects and closed VideoFrames
Fixed AudioSample rejecting SharedArrayBuffer data
Fixed missing validation for negative AudioSample channel counts
Fixed empty AudioSamples not being copyable
Fixed interlaced AVC being reported at half its height (#634)
Fixed HEVC SPS with inter-predicted short-term or long-term reference pictures being misparsed, giving wrong aspect ratios and colors
Fixed several AVC SPS parsing issues
Fixed VP9 RGB key frames (profiles 1 and 3) being misparsed, giving wrong codec strings
Fixed some AV1 codec strings having a wrong level, which browsers could reject
Fixed VP9 codec strings with level 0 when the file doesn't state a level, which browsers reject
Fixed AV1 hidden key frames being determined as key packets
Fixed avcC records of some AVC profiles (like 4:4:4 from x264) losing their extension fields
Fixed hvcC records putting prefix and suffix SEI into one array
HEVC is now encoded with hvc1 codec strings by default instead of hev1
HEVC codec strings read from ISOBMFF files now start with hvc1 or hev1 depending on the file's sample entry, instead of always hev1
Fixed missing color info for ProRes frames with a minimal frame header
Fixed ProRes video with an unknown color matrix being treated as RGB
Fixed 7.1 E-AC-3 being reported as 7 channels
Fixed 7.1 E-AC-3 muxed into MP4 being declared as 5.1
Fixed the way E-AC-3 is read from and written to ISOBMFF
Fixed ambisonic Opus with a demixing matrix (channel mapping family 3) losing the matrix in ISOBMFF
Fixed DTS-ES and DTS 96/24 being declared as plain DTS when written to ISOBMFF
Fixed canDecodeVideo() and canDecodeAudio() returning cached results for configs that only differ in their description
Fixed canDecodeVideo() ignoring the given size when choosing the codec string's level
Fixed canEncodeVideo() leaking its probe encoder in Firefox
AAC encoding now defaults to AAC-LC at all sample rates instead of requesting HE-AAC at 24 kHz and below
Fixed canDecode() not returning a promise for subtitle codecs
Fixed cover art stored in Vorbis comments being duplicated in FLAC to FLAC conversions
Conversion errors for tracks that can't be encoded now point to the DTS and TrueHD extension packages, and to @mediabunny/server in server-side environments
Fixed a conversion staying stuck in the executing state when starting it failed
Fixed canceling a conversion early throwing a generic error instead of ConversionCanceledError
Fixed unhandled promise rejections when writing a conversion's output failed
Fixed conversions reaching the 'done' state before their output had finalized, or even when finalizing failed
Fixed the index passed to conversion track option callbacks not matching the track number when only primary tracks are converted
Fixed zero-duration video frames inside the trim range being dropped when transcoding
Fixed trimmed copies of audio with delta packets (such as TrueHD) losing audio at the start
Fixed conversions discarding audio tracks instead of falling back to PCM when no compressed codec can be encoded
Fixed conversion progress being reported as NaN when there's nothing to convert
Fixed long gaps in audio allocating all of their silence padding at once, which could take gigabytes of memory
Fixed a broken error message for requested codecs the output format can't contain
Fixed conversions of implicitly signaled HE-AAC failing for formats like WAVE when the audio starts late
Fixed once event listeners being called twice during nested emits, and looping forever when re-registered while being emitted
Fixed EncodedPacket.clone() keeping the original byteLength when given new data
Fixed AES-128 encrypted HLS segments being kept in memory after a failed read
Fixed a WebVTT cue without a payload at the end of the text taking the whole text as its payload
Fixed WebVTT cues with hours above 99 being dropped
Fixed WebVTT text starting with a byte order mark failing to parse
@mediabunny/server: Decoding and encoding now use multiple threads where the codec supports it, making them faster
@mediabunny/aac-encoder, @mediabunny/flac-encoder, @mediabunny/mp3-encoder, @mediabunny/ac3, @mediabunny/dts, @mediabunny/truehd: Slightly faster encoding and decoding
@mediabunny/aac-encoder, @mediabunny/flac-encoder, @mediabunny/mp3-encoder, @mediabunny/ac3, @mediabunny/dts, @mediabunny/truehd: Fixed a memory leak in browsers and Deno with every worker created
@mediabunny/ac3, @mediabunny/dts, @mediabunny/truehd: Fixed a failure to create the codec worker breaking the extension for good instead of being retriable
@mediabunny/ac3: Fixed E-AC-3 with several syncframes per packet failing to decode
@mediabunny/ac3: Fixed the last decoded audio being dropped when flushing
@mediabunny/ac3: Fixed wrong decoded timestamps when the stream's sample rate differs from the track's
@mediabunny/ac3: Fixed decoding of streams with more than 8 channels
@mediabunny/ac3: Fixed the last 256 samples being lost when encoding
@mediabunny/ac3: Fixed 3 channels being encoded as 2.1, losing the third channel, and 4 channels as 4.0 instead of quad
@mediabunny/ac3: Fixed support being claimed for channel counts, sample rates and bitrates the encoder rejects
@mediabunny/ac3: Fixed slow encoding of very large samples
@mediabunny/dts: Fixed encoded audio being 512 samples late and losing its end; the first packet now starts 512 samples before the input to account for the encoder delay
@mediabunny/dts: Fixed wrong decoded timestamps when the stream's sample rate differs from the track's
@mediabunny/dts: Fixed decoding of streams with more than 8 channels, such as DTS:X
@mediabunny/dts: Fixed slow encoding of very large samples
@mediabunny/dts: Fixed the encoder not being reusable after a flush
@mediabunny/dts: Fixed float audio being truncated instead of rounded when encoding, which also overflowed at full scale
@mediabunny/dts: Fixed low quality settings being unsupported for multichannel audio; such bitrates now encode at the lowest bitrate DTS allows
@mediabunny/truehd: Fixed encoded streams shorter than 16 access units being unmuxable; they're now padded with silence
@mediabunny/truehd: Fixed wrong decoded timestamps when the stream's sample rate differs from the track's
@mediabunny/truehd: Fixed wrong decoded timestamps for packets holding several access units
@mediabunny/truehd: Fixed float audio being truncated instead of rounded to 24 bits when encoding
@mediabunny/aac-encoder: 7- and 8-channel audio is now encoded with channel configurations 11 and 12 instead of a program config element, which QuickTime silently dropped
@mediabunny/aac-encoder: Fixed 3 channels being encoded as 2.1, losing the third channel
@mediabunny/aac-encoder: Fixed 4 channels being encoded as 4.0 instead of quad
@mediabunny/aac-encoder: Fixed the encoder not being reusable after a flush
@mediabunny/aac-encoder: Fixed slow encoding of very large samples
@mediabunny/flac-encoder: Fixed the encoder not being reusable after a flush
@mediabunny/flac-encoder: Fixed float audio being truncated instead of rounded to 24 bits
@mediabunny/mp3-encoder: Fixed packet timestamps drifting for large start timestamps
@mediabunny/mp3-encoder: Fixed encoder errors producing garbage output instead of throwing
@mediabunny/mp3-encoder: Fixed failures to initialize the encoder going unnoticed
@mediabunny/prores: Fixed the pixel aspect ratio and color space set by the container being ignored
@mediabunny/prores: Fixed corrupted alpha in Safari for frames whose coded height exceeds their visible height
@mediabunny/prores: Fixed flushing resolving before the last frames were emitted
@mediabunny/prores: Fixed 4:3 and 16:9 aspect ratios in the frame header being treated as pixel aspect ratios
@mediabunny/server: 4-channel AAC can now be encoded as ADTS
@mediabunny/server: Fixed require('@mediabunny/server') throwing
@mediabunny/server: Fixed wrong decoded audio timestamps for some time bases
@mediabunny/server: Fixed a crash for planar audio with more than 8 channels
@mediabunny/server: Fixed VideoSample.transform() emptying the source frame, dropping encode options and the color matrix and range, and leaking memory when copying failed
@mediabunny/server: Fixed VideoSample.transform() ignoring alpha: 'discard'
@mediabunny/server: Fixed video with an unknown pixel aspect ratio throwing
@mediabunny/server: Fixed I420AP12 video samples failing to convert
@mediabunny/server: Fixed HEVC key frames forced close together being dropped
@mediabunny/server: Fixed VP8 encoding with alpha failing
@mediabunny/server: Fixed ProRes 4444 being encoded with 4:2:2 chroma subsampling
@mediabunny/server: Fixed an AVC quantizer of 0 switching to lossless encoding, which could break decoding
@mediabunny/server: Fixed precise video timestamps being lost or wrong after flushing
@mediabunny/server: Fixed some AV1 streams that browsers can decode failing to decode
@mediabunny/server: Fixed support being claimed for AAC, Opus, E-AC-3 and Vorbis channel counts and sample rates the encoders reject
@mediabunny/server: Fixed 3- and 5-channel AAC, AC-3 and E-AC-3 encoding failing
@mediabunny/server: Fixed 3- and 4-channel TrueHD being remixed instead of encoded losslessly
@mediabunny/server: Fixed encoded DTS being 512 samples late
@mediabunny/server: Fixed low quality settings being unsupported for DTS; such bitrates now encode at the lowest bitrate DTS allows
@mediabunny/server: Fixed AC-3, E-AC-3 and DTS encoding losing the end of the audio
@mediabunny/server: Fixed FLAC encoding reducing 24-bit and floating-point audio to 16 bits
@mediabunny/server: Fixed FLAC and TrueHD encoding adding silence to the end of the audio
@mediabunny/server: Fixed the video decoder ignoring the color space from the decoder config

mediabunny 1.66.0 v1.66.0 on Node.js NPM

mediabunny 1.66.0
v1.66.0

on Node.js NPM