🎵 Auditory Scene Analysis

Bregman's ASA: auditory streaming segregates concurrent sounds into streams. Frequency proximity (ΔF < ~semitone), harmonicity, onset synchrony, and spatial cues determine grouping.

SoundInteractive
Bregman ASA model · P pause · R reset · Increase frequency gap to hear streaming segregation

How it Works

The simulation models Bregman's auditory streaming phenomenon. Two interleaved tone sequences (A and B) alternate in time. The auditory system must decide whether to group them as a single integrated stream (heard as a galloping rhythm) or segregate them into two separate streams (each heard as a steady rhythm).

The segregation probability is computed from the frequency gap (ΔF in semitones), the presentation tempo, and the harmonicity of the tones. When ΔF exceeds ~3 semitones and tempo is fast, streaming occurs. The spectrogram view shows the frequency-time pattern as the auditory system analyzes it.

Segregation = σ(k · (ΔF − ΔF_crit) · tempo / tempo_crit)
ΔF in semitones: ΔF = 12 · log₂(f_B / f_A)
Stream B: f_B = f_A · 2^(gap/12)
Onset asynchrony cue: SOA = 1/tempo (seconds)

Frequently Asked Questions

What is Auditory Scene Analysis (ASA)?

Auditory Scene Analysis (ASA), developed by Albert Bregman, is the study of how the auditory system organizes the complex mixture of sounds reaching the ears into separate perceptual objects or 'streams.' Just as vision parses a visual scene into objects, hearing parses the acoustic scene into distinct sound sources.

What is auditory streaming?

Auditory streaming is the perceptual segregation of a rapid alternating tone sequence into separate streams based on frequency proximity. When two tones alternate rapidly and their frequencies are sufficiently different (more than a few semitones), they perceptually split into two separate streams — the 'galloping' illusion becomes two separate sequences.

What cues does the auditory system use for sound segregation?

The auditory system uses primitive (automatic) cues: frequency proximity (ΔF), harmonicity (is the sound harmonic?), onset synchrony (do components start together?), common amplitude modulation, spatial location (binaural cues), and spectral continuity. Schema-driven (learned) grouping uses knowledge of specific sounds like speech or music.

What is the cocktail party effect?

The cocktail party effect is the ability to selectively attend to one voice in a noisy environment with multiple speakers. It relies on ASA mechanisms: spatial separation, pitch differences between voices, rhythmic patterns, timbre differences, and top-down attention. This ability is impaired in hearing loss and some neurological conditions.

What is harmonicity in sound grouping?

Harmonicity refers to the property of having frequency components that are integer multiples of a fundamental frequency (harmonics). The auditory system tends to group harmonic components together as a single source. Mistuning a single partial by more than ~3% is sufficient to segregate it perceptually from the harmonic complex.

What is onset synchrony in ASA?

Onset synchrony is a powerful grouping cue: frequency components that start simultaneously tend to be grouped together as one sound, while those with different onsets segregate. Even a few milliseconds of onset difference can cause a component to pop out as a separate stream, exploiting the fact that a single sound source typically has synchronous components.

What is the role of the auditory cortex in ASA?

The auditory cortex, particularly the right hemisphere and areas in the superior temporal sulcus, plays a major role in ASA. Primary auditory cortex (A1) performs frequency analysis. Higher areas in the 'what' pathway process object identity, while the 'where' pathway handles spatial location — both critical for scene analysis.

How does frequency proximity affect streaming?

When two alternating tones are close in frequency (less than ~2-3 semitones), they tend to fuse into a single coherent stream perceived as a rhythm. When they differ by more than ~3-4 semitones, especially at fast tempos, they segregate into two separate streams, each with its own rhythm.

What is the role of attention in auditory streaming?

Attention modulates streaming: actively attending to one tone stream can make it stand out more and increase segregation. However, primitive (pre-attentive) streaming occurs automatically even without attention, as shown by the mismatch negativity (MMN) EEG response. Top-down attention adds to bottom-up primitive grouping.

What is the difference between primitive and schema-driven ASA?

Primitive ASA uses automatic, bottom-up acoustic cues (frequency, onset, harmonicity) that operate without learned knowledge of specific sounds. Schema-driven ASA uses top-down knowledge of specific sound patterns (e.g., a familiar voice, a word, a melody) to guide segregation. Both systems work together in real listening.

About this simulation

This simulator recreates Bregman's classic auditory streaming paradigm: two interleaved tone sequences, A and B, alternate at a controllable tempo and frequency gap, and a sigmoid segregation model — driven by the frequency gap in semitones, tempo, and harmonicity — decides whether your auditory system would hear one galloping rhythm or two separate steady streams. The Segregation % readout tracks that probability live as you adjust the controls.

🔬 What it shows

An event-driven visualization of alternating A/B tones plotted over time, alongside a live segregation probability computed from frequency distance (ΔF in semitones), presentation tempo, and harmonicity — the three primitive cues Bregman identified as driving stream formation.

🎮 How to use

Set Stream A frequency, Frequency gap (in semitones), Tempo, and Harmonicity, then switch Mode between Streaming Demo, Spectrogram View, and Grouping Cues to see the same underlying model from different angles. P pauses, R resets.

💡 Did you know?

The same two alternating tones can flip from "one galloping rhythm" to "two separate rhythms" purely by speeding up the tempo — segregation in this model scales with tempo because faster alternation gives the auditory system less time to bridge the frequency gap between successive tones.

Frequently asked questions

Why does increasing the Frequency gap push Segregation % toward 100?

segregationProbability() computes a sigmoid of (ΔF − 2.5) scaled by tempo and harmonicity factors, so once the semitone gap climbs past roughly 2.5 the raw score turns positive and the sigmoid output climbs quickly toward 1, matching Bregman's finding that streams reliably split once tones are more than a few semitones apart.

Why does raising Tempo also increase segregation, even with the same frequency gap?

The formula multiplies the frequency-gap term by tempo/4.0, so a faster tone rate amplifies whatever frequency separation exists — this captures the real phenomenon that fast alternating tones split into streams far more readily than the same tones played slowly.

What does the Harmonicity slider actually change in the model?

harmFactor = 0.5 + 0.5·harmonicity blends between 0.5 (fully inharmonic/noisy) and 1.0 (fully harmonic), scaling the segregation score — harmonic tone pairs are treated as more likely to belong to different sources when their frequencies diverge, mirroring how harmonicity helps the ear parse simultaneous sounds.

How is Stream B's frequency actually derived from Stream A?

freqB() computes params.freqA · 2^(gap/12), the standard equal-tempered semitone formula, so a gap of 12 always means exactly one octave above Stream A regardless of what frequency Stream A itself is set to.

Why does the segregation percept build up gradually rather than switching instantly?

The step() function updates streamIntegration by only 2% of the gap to its target value each frame, deliberately modeling the real perceptual finding that auditory streaming builds up over several seconds of listening rather than resolving on the very first tone pair.