Auditory Stream Segregation Task

1. Task Overview

This is an ABA-family adaptation: participants hear alternating low/high pure tones and report one or two perceived streams at the end of each sequence. It is not a replication of Bregman and Campbell's 1971 six-tone order-report experiment. Three frequency separations, two rates, and two exposure lengths are crossed. There is no objectively correct response or individual hearing norm. Endpoint duration comparisons are indirect evidence about buildup, not continuous perceptual-switch records. The small default sample does not yield a reliable individual segregation threshold.

2. Task Flow

Block-Level Flow

Trial-Level Flow

Controller Logic

3. Configuration Summary

Settings are from config/config.yaml. Run python main.py human, python main.py qa --config config/config_qa.yaml, or python main.py sim --config config/config_sampler_sim.yaml from the repository root. Install compatible PsychoPy, PsyFlow, numpy, pandas and PyYAML. QA and simulations use synthetic identity and fewer trials, without shortening any audio/stage timing.

a. Subject Info

b. Window Settings

c. Stimuli

d. Timing

e. Triggers

The generic report_hit / report_correct_keys framework fields indicate a registered valid key only; both keys are valid and neither is a correct percept. Analyze stream_count, two_stream, and missing_response, never interpret hit as perceptual accuracy.

All timing evidence is software timing. Native flip timestamps are not acoustic hardware onsets. WAV sampling can verify tone/frequency/silence parameters but cannot establish soundcard/headphone latency or ear-level SPL. Readiness is a participant confirmation, not a hardware test.

4. Methods (for academic publication)

Participants wear ordinary headphones in a quiet environment and confirm a sample is audible at a comfortable volume before proceeding. On each trial, a four-second silent fixation is followed by a diotic ABA- pure-tone sequence, then a five-second opportunity to report the percept at sequence end. Missing reports remain missing; no response is scored as correct. Two randomized blocks contain one repetition each of 12 factorial conditions. Chinese instruction and key assignments are identical across conditions.

The 1971 classic study provides the perceptual-grouping rationale but used six-tone order judgments. The present stimulus construction instead follows the ABA and endpoint-report family described by Dykstra et al. (2011), with original synthesized sounds. Fixed B frequency, 100-ms tones and 10-ms ramps, and the faster125-ms slot timing follow that method; selected gaps, slower rate, 4/12-second durations, trial count, response deadline, language, and digital level are documented adaptations. Bregman's cumulative-streaming study motivates the exposure and silent-separation factors. There is no adaptive threshold estimator. No psychophysical effect, clinical validity, hearing norm, acoustic calibration or continuous switch timing has been established by software validation.

See references/task_logic_audit.md, parameter/stimulus mappings and source links for the exact evidence/adaptation boundary. scripts/generate_audio.py reproduces the assets, with hashes/sample events in assets/manifest.json. Synthetic validation data are identified under validation/; they are not participant observations.