glint_audio_pure 0.6.2

Pure-Dart MPEG-1 Layer III (MP3) encoder and decoder — no native code, no FFI, runs on every platform including the web. Sibling of glint_audio.

0.6.2

• Shorten the package description to pub.dev's 180-character limit (it was 206, costing 10 pub points under "Provide a valid pubspec.yaml"), and mention the decoder — the package has exposed mp3Decode since 0.4.0 but the description still described it as an encoder only.
• No code changes; the encoder and decoder are byte-identical to 0.6.1.

0.6.1

• Robust decode of malformed input. mp3Decode no longer throws a RangeError (or hangs) on corrupt/truncated streams or non-MP3 files whose bytes happen to look like a frame sync: it resyncs past a bad frame and returns the cleanly-decodable prefix (possibly empty), never throwing on malformed content. Valid streams are unaffected. Fuzz-tested.

0.6.0

• Short/transient blocks now work in stereo and joint (M/S) too — the shortBlocks flag is on mp3EncodeStereo/mp3EncodeJointStereo as well as mono. Each channel schedules its own block-type chain; joint M/S combines the channels before the MDCT. Default stays OFF (byte-identical). A stereo transient reconstructs ~57–59 dB per channel, beating long-only.

0.5.0

• Short / transient block encoding (opt-in shortBlocks: true on mp3EncodeMono): a transient scheduler emits short/start/stop granules over attacks (long→start→short→stop→long), fixing pre-echo on percussive material. Default stays OFF — long-only output is byte-identical to 0.4.0. Fixes two window-switching quantizer bugs (ESC Huffman table selection for large coefficients; a missing anti-clip gain bound) so short blocks now reconstruct at ~70–78 dB and beat long-only on transients; the ffmpeg oracle agrees.
• Decoder now handles every block type (long / short / mixed / start / stop), so it decodes any real-world MP3, not just long-block streams.

0.4.0

• Decoder (mp3Decode): pure-Dart MPEG-1 Layer III decode (mono/stereo/ joint), so the codec now round-trips without any external tool. Matches ffmpeg's output bit-for-bit in quality.
• Joint (M/S) stereo encoding (mp3EncodeJointStereo).

0.3.0

• VBR streams now carry a Xing header (frame count + byte count + seek TOC), so players report exact duration and can seek.

0.2.0

• Stereo encoding (mp3EncodeStereo).
• Variable-bitrate (mp3EncodeMonoVbr / mp3EncodeStereoVbr, quality 0–9).
• Bit reservoir (main data spills across frame slots for better noise shaping).
• Huffman region optimizer (rate-optimal region/table selection).
• Quality: SNR 35 dB, NMR −7 dB on speech 128k (glint's own measure_audio.py).

0.1.0

• Initial release: pure-Dart MPEG-1 Layer III (MP3) encoder (mono, CBR).
• Full pipeline: subband analysis → MDCT (alias reduction + frequency inversion) → rate/distortion quantization with psychoacoustic noise shaping → Huffman → CBR frame assembly. No native code, no FFI, no runtime dependencies.
• Ported clean-room from the glint C++ codec suite.