Synthetic speech studies use controlled formant and F0 parameters to isolate cues

Evidence

Stimuli were synthetic vowels /a/ and /i/, generated using a MATLAB implementation of the Klatt formant synthesizer. Formant frequencies were {0.7, 1.0, 2.4, 3.3} kHz for /a/, and {0.3, 2.2, 3.0, 3.7} kHz for /i/. Stimuli were presented monaurally; either alone, or as “double vowels” /a, i/ (Fig. 1 ). F0 was either static (125 Hz, or 250 Hz for /a/, and 100 Hz, or 200 Hz for /i/), or sinusoidally modulated at 5 Hz, by ±2 semitones. Reverberation was added by convolution with real-room impulse responses recorded in a long corridor, at source-receiver distances of 0.32, 2.5, and 10 m (Tony Watkins; University of Reading, UK). We refer to these as “mild”, “moderate” and “strong” reverberation. For noise-masked vowels, Gaussian noise (5-kHz bandwidth) was added at signal-to-noise ratios of {10, 3, 0} dB. — Dijk, Pim van, Physiology, Psychoacoustics and…, 5031

Categorization

• DEFINE: Explains the standard stimulus construction (Klatt formant synthesizer, formant frequencies, F0 contours, SNR levels) used in speech-segregation research.