Investigation of the factors involved in the production of voiced and voiceless consonants in electric artificial larynx speech

Introduction: In speech using an electric larynx (EL), voiceless consonants are often mistaken for voiced consonants, affecting speech intelligibility (Kitajima, 1998). This challenge arises because it is difficult to create a voiceless segment or the difference in voice onset time (VOT) necessary to distinguish between voiced and voiceless consonants using the switch operation of the EL. This study investigates the factors necessary for the correct perception of voiceless consonants in EL speech through listening tests and acoustic analysis.

Method: Voice samples from an 80-year-old male, a seasoned laryngectomee using an electrolarynx for 11 years, were used. Meaningful words with Tokyo accent and voiced/unvoiced consonants (affricates /tɕi/ vs. /dʑi/) included: "chi" /tɕi/ and "ji" /dʑi/, as well as "ichi" /itɕi/ and "iji" /idʑi/. Each category had 10 items, totaling 80. In the listening experiment, 20 university students with no EL knowledge (naive listeners) determined if the 80 words sounded voiced or voiceless.Acoustic analysis used Praat software. For the speech samples judged to be voiced consonants by more than 70% of the participants and those judged to be voiceless consonants by more than 70%, we compared 1) burst noise intensity (noise occurring when the vocal tract opens during consonant production), 2) the duration of consonant closure and the duration of the consonant, and 3) vowel duration.

Results: Acoustic analysis was performed on 25 words judged in the listening results as voiceless and 15 as voiced. The acoustic analysis showed no significant differences in parameters for monosyllables. For disyllabic words, voiceless sounds had significantly stronger burst noise (p<0.05), and longer consonant closure (p<0.05) and consonant duration (p<0.01) compared to voiced sounds.

Discussion: The intensity of burst noise influenced the judgment of voiceless consonants, supporting previous research. The prolonged consonant closure and duration of the consonants in the speech samples judged to be voiceless were thought to be necessary to generate strong burst noise. From this, it was suggested that taking a sufficient duration of consonant closure and increasing oral pressure to create a burst might help in recognizing voiceless consonants. This could contribute to improving the intelligibility of EL speech.