Skip to content

v0.9.1

Choose a tag to compare

View all tags
@robmorgan robmorgan released this 02 Aug 02:17
v0.9.1
318b7fe
This commit was signed with the committer’s verified signature.
robmorgan Rob Morgan
SSH Key Fingerprint: bGTKOAt+CqU14/B6oEDtEiF0s5kNpjiyWCxOPjjdsHw
Verified
Learn about vigilant mode.

v0.9.1

What's Changed

Two additive analysis features aimed squarely at the DJ app: real broadcast-grade loudness metering for track autogain, and evidence-ranked tempo octave alternatives for one-tap BPM correction — now on crates.io.

Loudness Metering (BS.1770-4 / EBU R128)

  • measure_loudness(interleaved, channels, sample_rate) returns gated integrated loudness (LUFS), oversampled true peak (dBTP), and loudness range (LU) as a LoudnessMeasurement, with gain_db_to / gain_linear_to helpers for track autogain — via the pure-Rust ebur128 crate (#24)
  • Measured on the original interleaved channels, not the mono analysis downmix: BS.1770 sums per-channel energies, so a mid downmix reads up to ~3 dB low. analyze_for_dj (mono-only) therefore doesn't fill the field — callers measure and assign, as timestretch-cli analyze now does
  • PreAnalysisArtifact.loudness — schema v6, purely additive; v4/v5 sidecars stay compatible
  • Validated against BS.1770 reference points (full-scale 997 Hz stereo sine reads ~0.0 LUFS / 0 dBTP) and commercial masters: Music Sounds Better With You −8.9 LUFS / 1.0 dBTP, Cold Heart −7.9 LUFS / 2.1 dBTP
  • The simplified RMS estimate_lufs stays untouched as a benchmark-only A/B utility, so quality baselines don't shift

Ranked Tempo Candidates

  • The tempo tracker now scores the half/double metrical alternatives of its chosen Viterbi path with the same normalized tempogram-salience measure and exposes them ranked: TempoTrack::octave_saliences, BeatGrid::tempo_candidates, and PreAnalysisArtifact::tempo_candidates (schema v7, additive) (#25)
  • Enables a one-tap "halve/double BPM" correction ranked by measured evidence instead of blind ×2/÷2 buttons; alternatives outside the analyzed tempo range (50–220 BPM by default) are not offered
  • UX insurance, not an accuracy change — the corpus stays at 100% acc1 and detection behavior is untouched. Real-track check: 123.0 BPM committed at salience 1.000, half-tempo 61.5 offered at 0.612
  • timestretch-cli analyze prints the alternatives, plus METRIC lines with --verbose

Full Changelog: v0.9.0...v0.9.1

Assets 2
Loading