Acoustic models of co-varying vocal roughness and breathiness

Abstract

Dysphonia is characterized by several vocal qualities including breathiness and roughness. These qualities can coexist within a single voice. To examine the potential interaction between judgments of one voice quality in the presences of another, a large set of synthetic voices that co-varied along the breathiness and roughness continua were created and evaluated by 10 listeners. The perception of roughness (synthesized via amplitude modulation) was unaffected by degree of breathiness. The perception of breathiness, however, was affected by the degree of roughness: for low-breathy voices breathiness increased by about 2 dB as roughness increased from no roughness to maximum roughness; yet for high-breathy voices, breathiness decreased by about 2 dB as roughness increased. Current acoustic models of voice quality (pitch strength, auto-correlation peak, cepstral peak, partial loudness ratio, and glottal-noise-excitation ratio) do not explain the observed interaction of roughness and breathiness. To explain the observed interaction, we evaluate the possibility that roughness perception is dominated by temporal cues (which are unaffected by the presence of breathiness) and that breathiness perception is driven by spectro-temporal cues which are affected by the sub-harmonics generated by amplitude modulation. The sub-harmonics partially contribute to non-harmonic energy (increasing breathiness) and partially mask aperiodic energy (decreasing breathiness).