Hey Siri: How Effective are Common Voice Recognition Systems at Recognizing Dysphonic Voices?

Area of Science:

• Speech and Hearing Sciences
• Human-Computer Interaction
• Medical Technology

Background:

• Voice recognition technology is increasingly integrated into daily life.
• Individuals with voice disorders face challenges interacting with these systems.
• This can negatively impact their quality of life and independence.

Purpose of the Study:

• To evaluate the transcription accuracy of common voice recognition systems for dysphonic voices.
• To compare the performance of different voice recognition platforms (Apple iPhone 6S™, Apple iPhone 11 Pro™, Google Voice™).
• To assess the correlation between voice disorder severity and transcription accuracy.

Main Methods:

• A retrospective study involving participants with (n=30) and without (n=23) voice disorders.
• Participants read the "Rainbow Passage," and recordings were processed by voice recognition software.
• Word recognition scores were calculated and compared with auditory-perceptual and acoustic measures.

Main Results:

• Significantly lower mean word recognition scores were observed for dysphonic voices across all tested systems (68.6%-71.2%) compared to non-dysphonic voices (91.9%-93.8%).
• Strong negative correlations were found between the severity of dysphonia (CAPE-V ratings) and word recognition scores (R²=0.609-0.670).
• These correlations remained significant even after controlling for demographic and acoustic variables.

Conclusions:

• Current voice recognition systems exhibit poor transcription accuracy for individuals with dysphonic voices.
• Perceptual voice evaluation strongly correlates with the performance of these technologies.
• The findings underscore the necessity of considering the needs of patients with voice disorders in the development of voice-activated technologies.