Commit Graph
5 Commits
Author SHA1 Message Date
alehelandClaude Opus 5.5 7c652beec4 craft-audio: review fixes, round 1 (#446)
- `Encoder::with_application` checks the channel count before clamping
  the bitrate, so 86 channels or more return `Error::Channels` instead
  of panicking in `clamp`. Tested with 86 and 255.
- The ffmpeg oracle test skips a missing `opus`/`libopus` decoder
  through the testkit's `skip`, so `CRAFT_MEDIA_REQUIRE_ORACLES=1` fails
  it.
- README: the encode example pads at least the pre-skip after the audio
  before rounding up to a frame, so the encoder delay is flushed; the
  MP4 note says an explicit edit replaces the `media_start` edit and
  gives its fields.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 21:09:46 -05:00
alehelandClaude Opus 5.5 bdf8cb2a19 craft-audio: Opus encoder
EffectCraft's `effectcraft-opusenc` as `opus::encoder` (and
`opus::Encoder`) behind `opus-encode`: SILK, hybrid and CELT, 48 kHz,
20 ms frames, mono or stereo. The constructors and `encode_float`
return `Result` instead of panicking. `Bandwidth` is one type for the
decoder and the encoder, and the two `Mode` enums convert into each
other. `encode` now turns on `opus-encode`.

The encoder tests (modes, ffmpeg's two Opus decoders as oracles) and
`TrackDecoder` on Opus without an `OpusHead` come with it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 21:09:46 -05:00
Dylan M. Taylor 3f2897d2a6 craft-audio: test the SILK mid-only flag through a coded frame (#866)
The existing test sets the decoder's state by hand, so it also passes
with a mid-only flag kept from one frame to the next (the behaviour the
per-frame flag replaced). This one decodes a stereo frame that codes
the flag, then loses a frame: with a kept flag the side channel does
not restart and the test fails.
2026-11-11 17:26:31 -05:00
Dylan M. Taylor 44b4a23137 craft-audio, craft-media: the changes FilmCraft and SoundCraft made to their copies since extraction (#807)
* craft-audio: FilmCraft's FLAC encoder and MP1/MP2 decoder speed-ups

FLAC: per-order predictor kernels, merged Rice partition sums, the LPC
window kept per block length, and batches consumed by offset
(filmcraft 5eaff794, d9dad4a5). The output is unchanged: 64 golden cases
hash the same before and after.

MP1/MP2: the faster synthesis filterbank (filmcraft 51c755e5), with its
bit-exact test.

* craft-audio: FilmCraft's AAC and Opus decoder speed-ups and refactors

AAC: table-driven Huffman decoding, per-stage FFT twiddles and fused
stages, reused per-channel work buffers, and the encoder's memoised
per-band work (filmcraft f9627280, d587ceed, 6121635c, dc124ab7,
7a532eca).

Opus: radix-2/3/4 FFT kernels in the CELT IMDCT, the PVQ table built
once, one de-emphasis loop for both channels, the SILK resampler
without per-sample divisions and with eight partial sums, concealment
and FEC capped at 120 ms (filmcraft bf7ac420, 12e32026, c4dc64d1,
eb352ec8, 87b8e22d, 014b5e56, 78d77b72).

The per-frame mid-only flag this crate fixed in the SILK decoder is
kept in the refactored decode.

* craft-media: Ogg CRC-32 by slice-by-16

FilmCraft's 6763f9bf: sixteen independent lookups per 16 input bytes, so
opening a file is several times faster. Checked against a bit-at-a-time
reference on every short length and alignment, and across every split.

* craft-audio: SoundCraft's CAF cookie-size and FLAC vendor-length checks

A CAF magic cookie size above the signed range is refused instead of
passed to symphonia (soundcraft 3bd3965), and a Vorbis comment block
whose vendor string claims more than the block holds is refused
(soundcraft ddfe991). Each with its regression test.

* craft-audio: AC-3 reads bits through the shared BitReader

FilmCraft's 1c9c2e82: the AC-3 decoder's private bit-at-a-time reader is
replaced by craft-media's BitReader, with its two tests: the decoded
output of 2,000 synthetic and 12 real syncframes (and their damaged
variants) hashes the same as before, and the reader agrees with the old
one on random reads, skips and overruns. The three AC-3 features now
turn on craft-media's bitstream module.

* craft-audio: the quantiser rounds without a libm call

SoundCraft's d293336, the audio-io part: round-half-away by integer
arithmetic in the bounce's inner loop, with its test against f64::round
over the quantiser's range.
2026-11-11 11:32:33 -05:00
861c5e172d craft-audio: Opus decoder (7/10) (#445)
* craft-audio: Opus decoder

FilmCraft's `filmcraft-opus` decoder as `craft_audio::opus` behind
`opus-decode`: SILK, hybrid and CELT, every frame size, PLC and FEC,
multistream (`Decoder`) and single stream (`StreamDecoder`), 8-48 kHz
output.

- `TrackDecoder` decodes Opus; mono and stereo without an `OpusHead`
  (RTP, MPEG-TS) get a default one. Lost packets use Opus's own
  concealment.
- The native read path takes Ogg Opus (pre-skip and end trimming per
  RFC 7845), and Opus in Matroska/WebM and MP4, where packet starts
  after a seek come from the packets' own lengths.

The RFC 8251 conformance test runs all 36 vectors.
`scripts/fetch-opus-vectors.sh` downloads them (SHA-256 checked) and
the CI test job caches them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* craft-audio: review fixes, round 1 (#445)

- CELT concealment keeps its pitch in its own field. It reused postfilter_period_old, which the
  next good frame's post-filter cross-fades from.
- Matroska Opus seeks: a failed packet read no longer caches a wrong start; a one-byte packet
  is read as one byte.
- A malformed OpusHead in Ogg is reported as malformed instead of falling back.
- The libopus oracle tests skip through the testkit, so CRAFT_MEDIA_REQUIRE_ORACLES fails them.
- Decoder::reset documents that it re-arms pre-skip trimming.
- fetch-opus-vectors.sh retries its size probe.
- CI: the vectors are fetched after ffmpeg is installed, and apt's timeout runs as root so a
  stalled apt-get is killed and frees the dpkg lock.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* craft-audio: review fixes, round 2 (#445)

- CELT: the noise-based concealment sets skip_plc, as libopus does, so
  pitch-based concealment waits for two good frames in a row instead of
  extending history that holds noise.
- SILK: decode_only_middle is per call, as in libopus. A lost or mono
  frame no longer reuses the mid-only flag from an earlier frame.
- A damaged Opus packet is concealed at its own length (from its TOC and
  framing) rather than the previous packet's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Dylan M. Taylor <dylan@dylanmtaylor.com>
2026-11-11 07:36:57 -05:00