TL;DR
This paper investigates whether fundamental frequency (pitch) and formant space, crucial cues for speaker identification in normal hearing, are also effective in bone-conducted ultrasonic hearing (BCU hearing). The findings suggest that while both cues are utilized in BCU hearing, sensitivity to formant space is significantly reduced compared to air-conduction hearing.
What Problem Does This Paper Address?
Sensorineural hearing loss patients often struggle with speech comprehension, and conventional hearing aids can have limitations. Bone-conducted ultrasonic hearing (BCU hearing), facilitated by a BCU hearing aid (BCUHA), offers a potential alternative by transmitting speech sounds via ultrasonic vibrations through bone conduction. While the intelligibility of BCU hearing has been explored, its ability to convey paralinguistic information, such as speaker identity, remains less understood. Speaker discrimination is vital for effective communication, and this paper aims to determine if the acoustic cues typically used for this purpose – specifically pitch (fundamental frequency, Fo) and formant space – are preserved and utilized in BCU hearing. Understanding these cues is crucial for optimizing BCUHA technology for more natural and informative communication.
Methods
The researchers synthesized speech stimuli by manipulating two key acoustic features: fundamental frequency (Fo) and formant space. For Fo, three levels were used: original, 20% lower, and 20% higher than the original. Formant space was manipulated by stretching or shrinking it by 20%, reflecting changes in vocal tract length. This resulted in nine unique stimulus conditions (3 Fo levels x 3 formant space levels).
These manipulated stimuli were then presented to participants in two conditions:
- Air-Conduction (AC) Hearing: Stimuli were delivered through standard headphones.
- Bone-Conducted Ultrasonic (BCU) Hearing: Stimuli were presented as 30 kHz ultrasonic signals amplitude-modulated using the double sideband-transmitted carrier (DSB-TC) method. This modulated signal was delivered via a ceramic vibrator placed on the mastoid bone.
Participants were asked to judge whether pairs of presented words were spoken by the same or different speakers. The data was analyzed using multi-dimensional scaling (MDS) to visualize the perceptual relationships between the stimuli under both AC and BCU conditions.
Datasets
The original speech material was selected from "The Corpus of Spontaneous Japanese" (CSJ). Specifically, a short passage ("Uchu:") spoken by a female speaker was chosen. This speaker was selected to represent an average vocal profile from a large dataset of female speakers within the CSJ. No new datasets were introduced; existing corpora were used for stimulus selection.
Key Insights and Ideas
- BCU Hearing Utilizes Speaker Cues: The research confirms that BCU hearing, like normal air-conduction hearing, allows listeners to discriminate speakers based on changes in both pitch (Fo) and formant space. This is a significant finding for the potential of BCUHA in conveying more than just linguistic content.
- Formant Space Sensitivity is Reduced in BCU: A key takeaway is that listeners are less sensitive to variations in formant space when hearing through BCU compared to AC. This suggests that the physical characteristics of the vocal tract, which are strongly reflected in formant space, are not as clearly perceived via bone conduction.
- Pitch Remains a Viable Cue: While formant space sensitivity decreases, pitch (Fo) appears to remain a relatively strong cue for speaker discrimination in BCU hearing, contributing significantly to the perceptual space.
- MDS Reveals Perceptual Structures: The use of multi-dimensional scaling effectively visualized how listeners group and differentiate the synthesized stimuli, providing a clear comparison of perceptual organization under AC and BCU conditions.
Key Results
- In AC hearing, formant space was found to be the dominant factor for speaker discrimination, with clear clustering of stimuli based on formant space size. Pitch also played a role, but to a lesser extent.
- In BCU hearing, the sharp clustering observed in AC was absent. However, the results indicated that both Fo and formant space contributed to speaker discrimination. Dimension 1 of the MDS plot primarily reflected Fo, while Dimension 2 reflected formant space, suggesting a more balanced, albeit less distinct, contribution of both cues.
- The study found that identical stimuli (same Fo and formant space) were consistently perceived as the same speaker's voice in both AC and BCU conditions, indicating the reliability of the synthesized stimuli.
Limitations and Future Work
- Reduced Formant Space Sensitivity: The primary limitation identified is the reduced sensitivity to formant space in BCU hearing. The authors suggest this may be due to a lower frequency resolution in BCU hearing, particularly at lower and higher frequencies, which could impact the perception of formant differences.
- Further Investigation Needed: The paper calls for further research to address the reduced sensitivity to formant space. This could involve exploring methods to improve frequency resolution in BCU hearing or investigating other acoustic cues that might be more robustly transmitted via bone conduction for speaker identification.
- Stimulus Set: While the study used synthesized stimuli to control specific acoustic features, future work could explore speaker discrimination using more naturalistic speech samples to assess the real-world performance of BCU hearing.
Summary updated