Release v0.1.3 - Various Improvements · Audiflo/flo

Added

Block switching in the lossy encoder: the MDCT path now switches between long and short windows behind a transient/loudness-spike detector, reducing pre-echo and improving quality on transient-heavy material.
Random access: a dedicated seeking module exposes a table of contents (get_toc), frame-accurate decode (decode_frame_at), and time-based seek (seek_to_time) over encoded bytes.
Analysis API: waveform peaks and RMS, spectral fingerprint and dominant-frequency extraction, and spectral similarity, all pure functions over encoded bytes. Surface in the demo File Info card and the CLI analysis subcommand.
EBU R128 loudness: integrated loudness and true-peak measurement computed during encode and reported on decode.
ALAC decode support in reflo, via Symphonia.
FFT backend abstraction: a new dsp module with swappable fft-rustfft and fft-microfft backends.
no_std: libflo-audio compiles with --no-default-features for embedded targets (thumbv8m.main-none-eabihf); reflo follows with a conditional std feature.
CLI: reflo-cli splits into its own crate (published to crates.io, cargo install reflo-cli gives the flo binary) and gains metadata, analysis, and validate subcommands alongside encode, decode, and info.
Web demo: the desktop demo gains waveform and spectral visualization, worker-thread decoding, and TOC-based seeking during playback.
Documentation: docs move to a generated docmd site with new file-format, no_std, CLI reference, and architecture guides.
Release infrastructure: OIDC trusted publishing for both crates.io and npm, so no registry tokens are stored in CI; npm publications carry SLSA provenance and the CLI/WASM artifacts get build-provenance attestations. Release workflows (with dry-run for manual testing) split into release-cargo.yml and release-npm.yml.
Security gates: cargo-audit (advisories) and cargo-deny (licenses, bans) run in CI on lockfile changes and on a weekly schedule, with read-only permissions.

Changed

Dependencies: bumped symphonia 0.5 -> 0.6.1 and migrated the reflo decode/metadata pipelines to the new API (AudioCodecParameters,
AudioDecoderOptions, metadata revisions, probe-by-value options). All Rust workspace deps updated to their latest patch/minor versions (serde, serde_json, serde_bytes, messagepack-serde, serde-wasm-bindgen, rustfft, blake3, wasm-bindgen, js-sys, web-sys, anyhow, chrono, clap).
Workspace layout: crates consolidate into a single Build/ workspace (crates/libflo, crates/reflo, crates/reflo-cli, crates/flo-fixtures) with shared Python scripts/ and a Justfile command layer.
npm scope: the JS packages move from @flo-audio to the @audiflo org (@flo-audio/libflo-audio -> @audiflo/libflo, @flo-audio/reflo -> @audiflo/reflo); the old package names are deprecated.
Streaming encoder: adds a compression-level knob, chunked flush_into_frames, and an explicit finalize step that writes the trailing metadata and footer.
Demo downloads: flo files are offered as inline downloads with the audio/x-flo MIME type instead of generic application/octet-stream.
Test restructure: native Rust integration tests now live in Tests/ (Tests/libflo/, Tests/reflo/) and are wired via [[test]] paths in the crate manifests; Jest WASM tests moved to Tests/libflo/js/.
Fixtures: added Build/crates/flo-fixtures, an in-code signal synthesizer that replaces sox-based generation. just examples now writes .wav/.flo files to Tests/fixtures/ and regenerates Examples/.
Tooling: .devcontainer/ dev container (Rust + Node 24 + Python 3.14, wasm32 target, just, cargo-deny, wasm-pack) and toolchain version pins: .nvmrc (Node 24), .python-version (3.14), and an .envrc for direnv users.

Fixed

True lossless: the lossless path no longer loses data; encode then decode round-trips bit-exactly. The decoder stays compatible with the released v0.1.2 format (which coded mid = L + R).
Duration bookkeeping: sample counts across read, write, and seeking are now per-channel sample-frames rather than interleaved samples, fixing duration and total-frame reporting.
Wasm encoder: next_frame returns null (not undefined) on end-of-stream; flush() now runs the final partial frame through flush_into_frames.
ALPC serialize_channel serialization layout (order, coeffs, shift bits, residual encoding, Rice parameter) so encoded frames round-trip correctly.