No recording

Reading the sound plane

The object types follow Denis Smalley's spectromorphology. Each object sits where its energy lies in the spectrum; shape and fill indicate the type, brightness the level.

Impulse

Attack-impulse without continuant: vertical line across the excited band; in detail view with a faint decay.

Block

Attack-continuant: a sound mass with common onset and offset spanning at least half an octave. Solid: stationary. Striped: amplitude-modulated. Small square: under 120 ms. Detected partials are drawn as fine lines.

Tone

Narrow-band, stable spectrum (note). Line: sine tone. Connected lines: tone complex with common onset. Curve: glissando. Dashed: held tone over 1.5 s.

Field

Stationary texture (Smalley's texture as opposed to gesture). The outline shows the occupied band over time, the dot density the spectral flatness: dotted = noisy, hatched = tonal or filtered.

Psychoacoustic lanes

Loudness N (sone)

Zwicker model on third-octave spectra with simplified masking slope, 46 ms window. Twice the sone is perceived as twice as loud. Calibration: 1 kHz at 70 dB SPL = 8 sone; 0 dBFS sine is assumed = 90 dB SPL.

Sharpness S (acum)

DIN 45692 from specific loudness. Share of high bands: bright, pointed, sharp.

Roughness R (asper)

Simplified Daniel & Weber model: 24 Bark bands, envelope modulation 15–300 Hz with maximum at 70 Hz, correlation of adjacent bands. 200 ms window. Calibration: 1 kHz, 60 dB, 100 % AM at 70 Hz = 1 asper.

Spectral centroid SC

Energy-weighted mean frequency; correlates with perceived brightness. Drawn on the sound plane as a band of ±1 standard deviation on log frequency.

Flatness

Wiener entropy. Near 0: line spectrum, tonal. Towards 1: noise.

Spectral flux

Positive spectral change per frame. Measure of motion and eventfulness.

Event density

Onsets per second per object type. Solid: impulses. Dashed: blocks. Dotted: tones.

Stereo width

Side energy / (mid + side). 0 = mono, 0.5 = uncorrelated.

Chords

Chroma (pitch-class energy) from spectral peaks 65 Hz–5 kHz, matched against major, minor, 7, maj7, m7, diminished and sus4 templates; bass below 300 Hz favours the root; Viterbi smoothing against rapid changes. Shown only where the spectrum is tonal and a template fits clearly. Inversions, extensions and enharmonic spelling are not analysed (roots: C C# D Eb E F F# G Ab A Bb B).

Novelty

Foote novelty from the self-similarity matrix. Form resolution sets the kernel width: fine finds short sections, coarse only the major caesuras. Digital silence longer than 1 s always counts as a boundary.

Limits of the method

Object detection is automatic: harmonic/percussive separation by median filtering, sound masses grouped by common onset, tones tracked via spectral peaks with at least 18 dB prominence. Dense, reverberant or instrumental music produces more misclassifications than clearly separated electronic sounds.

Loudness, sharpness and roughness approximate the standard methods at an assumed listening level. They correlate closely with the reference implementations; for form analysis, contours and relations matter, not absolute values.

Long recordings need time and memory: a 10-minute piece takes about half a minute. Beyond 40 minutes, weaker devices may run short. Multichannel recordings are reduced to stereo.

Reading the pitch plane

Vertical axis: pitch in semitones (E1–E7), grid lines at every C. Background: semitone spectrogram (MIDI 24–107, 4096-point FFT).

Melody

Bars: melody notes, rounded to the semitone. Thin line: continuous pitch contour (10-cent resolution).

Beats

Thin vertical lines: beats. Strong lines: first beat of a bar. Drawn only where the tempo is clear.

Chords

Chord lane: symbol (root, quality) and Roman numeral relative to the local key.

Key

Key lane: local key areas; empty where no key profile fits clearly.

Methods

Tempo

Onset envelope from positive band-energy change; autocorrelation in 8 s windows every 1 s with a log-normal preference around 110 BPM; Viterbi over 40–220 BPM.

Beats and bars

Beats by dynamic programming on the onset envelope (Ellis 2007). Bar grouping 2, 3 or 4 from the periodicity of accents (low-register onsets, chord changes), determined separately for each stretch of steady tempo.

Key

Krumhansl–Kessler profiles correlated with the chroma in 12 s windows; Viterbi against frequent changes; each boundary moved to the chord change that best separates the two keys.

Roman numerals

Chromatic degree of the chord root relative to the local key; upper case = major, lower case = minor, ° = diminished. Minor keys counted from natural minor (III, VI, VII), the leading-tone chord as vii°.

Melody

Harmonic summation of spectral peaks (8 harmonics, weight 0.8ⁿ), preference for the higher voice, Viterbi over the strongest candidates; notes by semitone segmentation, repeated notes split at re-attacks.

Onset density

Peaks of the onset envelope per second (1 s window).

Limits of the method

The melody is the most prominent, usually highest voice. In dense polyphony or orchestral tutti it may switch between voices; octave errors are possible.

Tempo can be estimated at half or double speed. Bar grouping says how many beats form a bar, not the notated time signature; downbeats are omitted where the tempo is unclear.

Key and Roman numerals assume major/minor tonality. Modal, chromatic or atonal music yields few or no key areas.