Fusion: reproducing Bregman & Ahad

Listen for one rich tone at first, then the harmonic that goes off and on: each time it comes back it stands out as a pure tone of its own for a moment, before it fuses again. The demonstration does this for each of the first ten harmonics in turn, so the tone that stands out rises in steps of 200 Hz.

Houtsma, Rossing & Wagenaars, Demonstration 1, after Plomp (1964). A complex tone of 20 harmonics of 200 Hz, then for each of harmonics 1 to 10 in turn, the harmonic switched off and on again. Their booklet reports that while the harmonics stay steady the tone is heard as a whole, and that a harmonic turned off and on stands out clearly. It gives neither the harmonics' levels nor the timing; here the harmonics are equal in amplitude, and each harmonic gets 1 s with every harmonic on, then twice 0.5 s off and 0.5 s on, with 10 ms ramps.

Listen for a thin pure tone that begins to stand apart from the rich tone after a few steps of the first series, and fuses back into it during the last few steps of the second. Counting the steps it takes gives the mistuning, in percent, at which you start to hear it.

Bregman & Ahad, Demonstration 18, after Moore, Glasberg & Peters (1986). A tone of the first eight harmonics at equal amplitude in sine phase, 250 ms long with a 10 ms rise and a 9 ms decay, and 1 s of silence between presentations, as in the booklet. The third harmonic is the target: it stays at its frequency while the other seven, the harmonic frame, move. In the first series the target plays alone twice, then the full tone in tune, then the frame drops by 1 % of its starting frequency at each presentation for 15 steps. In the second, after the target alone twice, the frame starts 15 % low and rises 1 % at a time back into tune. The booklet reports the target heard as a separate pure tone after a few steps of mistuning, and that holding it still while the frame moves keeps a changing frequency from drawing attention to it. The fundamental is not in the booklet's text; 200 Hz, a 600 Hz target, is ours.

Listen for one steady tone, then two sounds while some of its harmonics rise and fall: a steady tone and a warbling one, the steady one a little thinner than before. When the harmonics come back together they are heard as one tone again.

Bregman & Ahad, Demonstration 19. Harmonics 1 to 8 of 400 Hz at equal amplitude in sine phase, 6.2 s with 40 ms rise and decay. After 2 s steady, harmonics 2, 4 and 8 follow a fundamental that moves in straight lines up to 600 Hz and down to 300 Hz, four times, each rise and fall 0.25 s, then back to 400 Hz, while harmonics 1, 3, 5, 6 and 7 stay put, all as in the booklet. The booklet reports the two sets heard as separate sounds while they move apart and as one tone again when they come back together. It plays the pattern twice; here it plays once.

Listen for one tone sliding up and down, quickly and then slowly. In the third and fourth sounds the two sets move at different rates, and the booklet reports two distinct tones, one moving slowly and one quickly. In the fourth the rates are swapped between the sets, so the tone that sounded pure now moves slowly and the rich one quickly.

Bregman & Ahad, Demonstration 20. Two sets of partials, equal in level and starting in sine phase: set 1 holds three partials in the ratio 3:4:5, set 2 a single partial, the fundamental itself. The slow pattern rises from 400 to 800 Hz in 1.25 s and falls back in 1.25 s; the fast one moves between 400 and 660 Hz five times, each rise and fall 0.25 s. Each sound lasts 2.5 s with 40 ms rise and decay. First both sets on the fast pattern, then both on the slow one, so that together they are harmonics 1, 3, 4 and 5 of one moving fundamental; then set 1 on the slow pattern and set 2 on the fast one, as in the booklet's technical details. The booklet plays that third sound twice; this reproduction plays it once and then swaps the rates, set 1 fast and set 2 slow. The booklet's text describes the slow tone as pure and the fast one as rich, which is the fourth sound here, while its technical details describe the third. The pauses between sounds (0.5 s) are ours.

Listen for a pure tone, which seems to carry on when a steady rich tone joins it; then, when every harmonic starts to wobble together, the pure tone disappears into the sound and a singing voice on a vowel emerges. There are three examples, for three vowels on three pitches; this first one is "oh" on 400 Hz.

After Bregman & Ahad, Demonstration 24, which plays examples made by John Chowning with FM synthesis. For each vowel, a pure tone, then the same tone joined by its harmonics at the levels they would have in a vowel, then all of them modulated by one pattern of vibrato and random micromodulation. The pure tones are 400, 500 and 600 Hz, and the vowels "oh" (more), "ah" (bomb) and "e" (bed), as in the booklet. The booklet reports that the pure tone is heard continuing into the steady rich tone (the old-plus-new principle, the subject of the demonstrations on capture), and that once the harmonics move together the tone can no longer be heard on its own and a voice emerges. These are not Chowning's tones. Ours: the harmonic levels come from three resonances at Peterson & Barney's (1952) average formants for women, for "aw" (closest to "oh" as in more), "ah" and "eh", with bandwidths 80, 90 and 150 Hz, on a source falling 6 dB an octave, with the fundamental kept at the pure tone's level; the vibrato is 5 to 6 Hz and ±2 %, plus random micromodulation of 0.5 % RMS below 8 Hz. The booklet's timing is 5 s of pure tone, 5 s of steady spectrum and 10 s of modulation; here it is 3, 3 and 6 s.

Listen for the same three stages: a pure tone, the tone joined by a steady rich tone, then a voice. This one is "ah" on 500 Hz.

After Bregman & Ahad, Demonstration 24, its second example, built as the first.

Listen for the same three stages again, now "e" (as in bed) on 600 Hz.

After Bregman & Ahad, Demonstration 24, its third example, built as the first.

Listen for a chord of three pure tones, then a chord of three steady rich tones, then three voices singing three vowels on three pitches. The booklet expects most listeners to tell at least two of the vowels apart.

After Bregman & Ahad, Demonstration 24, its fourth part: the three examples above played at once, each with its own pattern of vibrato and micromodulation, as in the booklet. Here the voices sing for 4 s rather than 6 s.

Listen for the order of the two middle notes in each pair of clusters: is it the same in both clusters, or different? The booklet reports that as the notes start more abruptly they sound clearer and more bell-like, and their order gets easier to hear. The answers for this reproduction, pair by pair: different, same, different, different.

Bregman & Ahad, Demonstration 21, after Bregman, Ahad & Kim (1994). Clusters of four pure tones, each starting 120 ms after the one before and all ending together; the notes are L = 700, M = 800 and H = 900 Hz, in the order MHLM or MLHM, so that both clusters start and end on M, the positions that are easiest to hear. Each tone's envelope rises in a straight line and falls in a straight line with no pause, and the whole of the first tone takes 1 s; the later tones rise at the same rate but stop where they meet the first tone's falling line and follow it down, so no later tone is ever louder than an earlier one (with a 125 ms rise their peaks are 0.88, 0.76 and 0.64 of the first, as in the booklet). The first tone's rise is 250, 125, 50 and 10 ms in successive pairs. Which pairs are the same is ours, as are the silences: 0.5 s within a pair and 1.5 s between pairs.

Listen for an even beat of soft tones, twice a second. When irregular tones of the same pitch are mixed in, the beat is hard to follow. When each irregular tone gets four partners an octave or two away, the booklet reports that they fuse into rich tones and the even beat of pure tones can be heard again; when the partners start and stop slightly late, the fusion fails and the beat is lost again.

Bregman & Ahad, Demonstration 22, rhythmic masking release. Targets and maskers are 1000 Hz tones with 5 ms rise and decay and equal amplitude; the targets are 50 ms long with a regular 0.5 s from onset to onset, and the flankers, partners of the maskers, are 250, 500, 2000 and 4000 Hz tones each 12 dB below the masker, all as in the booklet. Four parts: ten targets alone, then with the maskers, then with maskers and flankers, then with the flankers delayed. Ours: one masker in each gap between targets, at a random time (seed 22), and 30 to 60 ms long, the range the booklet gives for the flankers; the delay of the flankers, 30 ms at onset and offset; and each part played once, where the booklet plays the middle two twice.

Listen for a steady tone that comes and goes in a fluctuating hum, getting softer at every beep. Count how many beeps you hear in each of the three parts. Hall, Haggard & Fernandes (1984) found the tone detectable at lower levels when the noise's fluctuations were shared by bands far from the tone, and Bregman's booklet explains the effect as the shared fluctuation telling hearing that the bands are one sound, which leaves the steady tone apart.

Comodulation masking release, after Hall, Haggard & Fernandes (1984), in the form with separate flanking bands that later work made common (see Moore, 1990); it is on neither disc, and every number here is ours. A 1000 Hz tone, 300 ms long with 20 ms ramps, every 0.8 s, stepping down 3 dB at each beep from the level of the noise band it sits in. The noise is a 1000 Hz carrier times a slowly fluctuating envelope (low-pass Gaussian noise below 25 Hz), a band about 50 Hz wide. Three parts: that band alone; with four more bands at 500, 700, 1400 and 2000 Hz, all with the same envelope (comodulated); and with the same four bands, each with an envelope of its own.

Listen for words. Three gliding whistles: can you tell what the speaker is saying? The booklet reports that after a little practice listeners can understand sine-wave speech, sentences more easily than single words, although the three tones share no fundamental and do not start or move together the way the harmonics of a voice do. The answer is folded away below, under "Show the answer".

A stand-in for Bregman & Ahad, Demonstration 23, after Remez, Rubin, Pisoni & Carrell (1981). The disc plays sine-wave speech made from parameters supplied by Robert Remez and Philip Rubin, which are not ours to use. This reproduction makes it from the male talker's recorded sentence (CMU ARCTIC, speaker bdl): so.formant_track follows the first three formants, and each becomes a pure tone that glides along it, at levels that follow the sentence's short-time RMS at 0, -6 and -12 dB. The tones are switched off where so.f0_track finds no voicing (or a formant track has a gap); the booklet's examples instead put steady tones in noisy stretches such as "s", so this one leaves gaps there. The levels are ours.