Perception of concurrent sentences: Within and across-formant grouping in harmonic and frequency-shifted speech

Keyword identification in one of two simultaneous sentences is substantially improved when they differ in fundamental frequency (F0); this effect is greatest for almost continuously voiced speech. ΔF0 may act by improving voice identification better first-formant (F1) definition or better across-formant grouping or voice tracking. Sentences were monotonized and resynthesized to give a range of ΔF0s (0–10 semitones; F0s = 90–160 Hz). Sentences were additionally resynthesized after applying a frequency shift of 25% of F0 to the monotonized excitation source—making it inharmonic but with regularly spaced components—while preserving the original formant frequencies. Sentence pairs were created by embedding shorter targets within longer interferers. The large improvement with increasing ΔF0 found for harmonic sentences was reduced but still substantial for frequency-shifted sentences. In both cases, swapping target and interferer F0s across spectral regions (below vs above 800 Hz, F1 vs higher formants) caused substantial intelligibility loss only for large ΔF0s, indicating an important effect of changes in across-formant grouping. Listeners made few target-tracking errors, but these errors were more frequent for smaller ΔF0s and for stimuli with more ambiguous pitches. The results extend the range of perceptual phenomena usually attributed to harmonic processing to grouping by spectral regularity.