Subharmonics in Normal Voice: An Acoustic Database Study

Objective: While subharmonic phonation is commonly observed in disordered voice, speakers without any voice concern can also produce a voice with subharmonics both voluntarily and involuntarily. This leads to the importance of establishing a statistical baseline for subharmonics-specific acoustic parameters among normal voice population. This presentation reports the findings from our acoustic analysis of the normal voice samples from four voice databases.

Methods: 928 acoustic recordings of sustained /ɑ/ were gathered from Perceptual Voice Quality Database, KayPENTAX Voice Disorder Database, Saarbrueken Voice Database, and Universidad Politécnica de Madrid (UPM) Database. Each recording was first resampled at 4000 samples/second and was split into 50-millisecond analysis segments (40408 total). A subharmonic detection algorithm modeled the acoustic waveform of each segment as a sum of time-varying (sub)harmonics. Up to period-tripling subharmonics were detected. Three per-segment subharmonics-specific parameters were computed from each model: subharmonic period, subharmonics-to-noise ratio (SHR, the ratio of the total average powers of subharmonic tones and of harmonic tones), and subharmonic richness factor (SRF, the ratio of the number of subharmonic tones detected and the maximum possible). We considered subharmonics at SHR = –20 dB to be barely perceivable. Both per-segment and per-recording statistical distributions were studied.

Results and Conclusions: The presence of short subharmonic bursts was not uncommon although strong ones were rare for normal voices. The algorithm detected potential subharmonic activities in 26% of the segments and 96% of the recordings, indicating the highly sensitive nature of the detector. By limiting the subharmonics to those with SHR > –20 dB, the occurrence rate reduced to 4.7% of the segments with –14-dB mean SHR and 73% mean SRF. Of those segments, 70% were period-doubling and 30% period-tripling. Of the recordings, 40% had SHR > –20 dB. Among them, the mean duration of the subharmonic bursts was 120 milliseconds with the longest of 2.5 seconds. The 95-percentile of the SHR was –12 dB. These findings are crucial to establish the point of reference towards assessing the presence of subharmonics in disordered voices.