Learned acoustic cues, individualized auditory training and retained speech recognition in noise.Open the stimulus workspace ↓
FIG. 01MI-STM · personal anchor trajectories
SCHEMATIC BANK · NO MI RUN
REQUESTED POSITION
0.32/0.24/+4.8dB
Gray cloud · competing soundIllustrative field, not measured spectra
SPEECH STRUCTURE
X changes both harmonic amplitudes and broadband consonant spectra; F0 and harmonic spacing stay fixed. Y changes syllable durations and pauses. Below: harmonic /ɑ/ and aperiodic /s/ spectra, compared with X = 0.
Model and sources
Every curve is drawn from phonetic parameters; no photograph or recorded spectrogram is stretched. Harmonic frequencies remain at n × F0, with F0 fixed to the reference median. X changes resonance centres, bandwidths and gains independently, plus partial-specific amplitudes. Harmonic energy is normalized at each voiced phase; increasing X does not raise the whole spectrum or its total energy. Unvoiced /s, ʃ/ frication and /p, k, t/ release bursts occupy continuous frequency bands, independently of F0. Their spectral envelopes also redistribute energy with X; they belong to the target speech, not the external masker. Each source spectrum is normalized separately, without claiming equal perceived loudness. These amplitude patterns are illustrative, not measured spectral-distance calibrations. The F0 tolerance controls bank eligibility, not this drawing. Six illustrative consonant–vowel events retain their order while their durations and pauses change on a fixed 3-second axis. The drawing is a source–filter illustration, not measured bank audio or an inverse R³/T³ model; it does not establish intelligibility.
F · familiar referencePoints · stored candidates at each masker levelDRAG TO ROTATE ↔
3D rendering is unavailable in this browser. The parameter controls and anchor metadata remain available.
PERSONAL TRAJECTORY · F0-MATCHED ANCHORSSelect anchors with increasing R³ and T³ distance.Discrete recordings · masker level held along the route
BANK STATUSIllustrative dataset · 1,024 candidate coordinates
＋
Generate → measure → select. X/Y select a pool anchor relative to a familiar voice. Z changes the scene after selection. Acoustic distance is not perceptual difficulty.
All candidates · measured output coordinates · one matched-content slice