Ratings of Intelligibility, Naturalness, and Listener Effort for Mask Acoustic Filtered Speech Produced Using Clear Speech Strategies

Abstract

The use of face masks affects both the acoustic and visual information conveyed byspeech, especially in the presence of competing noise. The masks obscure the nose and mouth, obstruct visual cues that contribute to good speech intelligibility (Erber, 1969; Ewertsen, 1971, Llamas, 2008). In addition, each type of face mask acts as a unique filter on the individual talker’s speech acoustics by attenuating speech in the range of 2000-7000 Hz, a frequency range that is important to speech understanding (Goldin et al., 2020). Previous work has found 1-17% decreases in speech recognition accuracy with different mask types (Radonovich, 2009). The variance appears to be related to talker, speech material, and environment. There are a number listener-oriented changes in speech production style that can be utilized by a talker to enhance speech intelligibility and ultimately improve the listener’s comprehension of the message. Clear speech differs acoustically from conversational speech in several ways. According to Schum, “Clear speech is characterized as the full, accurate, and precise production of each speech sound in each word. Each word in a sentence is fully produced. Word endings are not dropped. Natural voice stress and phrasing are retained. Sentences are not produced in a staccato, halting, exaggerated, or otherwise unnatural manner” (1997, p. 38). Some of these clear speech strategies include slowing rate, overarticulating, increasing loudness, and use of clear speech as compared to conversational speech. These strategies result in enhanced phonemic contrasts and have been used to counter a variety of communication barriers (e.g., Park, Theodoros, Finch, Cardell, 2016). Clear speech strategies may improve audibility or speech understanding to overcome communication barriers such as social distancing and mask wearing, strategies recommended by the Centers for Disease Control (CDC) to mitigate the spread of COVID-19. For example, over-articulation and a slowed speech rate lead to vowel space expansion, which increases the F1 and F2 contrasts between vowels, and reduced target undershoot, which results in better approximated vowel production (Bradlow, 2002). Clear speech is more intelligible than conversational speech for a wide range of listeners and conditions [for a review of the clear speech research, see Smiljanic and Bradlow (2009); Smiljanic (2021); Uchanski (2005)]. For listeners with hearing loss, clear speech has been found to improve conversational 7 fluency when compared with conversational speech (Caissie, 2005). As such, clear speech strategies are a useful aural rehabilitation tool and have been recommended for conversation partners of those with hearing loss to increase conversational fluency (Beyer, 2000; Caissie, 2010; Krause, 2002). It is unclear how different clear speech strategies affect understanding of speech when combined with different types of masks. The purpose of this audiology doctoral project was to extend our understanding of the interactions between acoustic filtering from two different types of face masks (simple surgical type mask and N95 respirator) and four different talker strategies (clear, loud, slow, over-articulated) in the domains of intelligibility, naturalness, and listener effort to improve intelligibility for native adult speakers of American English. This perceptual listening experiment builds upon previous works to provide a deeper understanding of how talkers can adjust their speech production style to affect listener perception when the talker is wearing a mask (Swink, 2017). A perceptual listening experiment used acoustic mask filters applied to pre-recorded male and female speech samples. Results revealed that across listeners intelligibility and naturalness were not substantially affected by the different mask conditions compared to no mask. However, for all speaking conditions except over-articulation listener effort increased with simple surgical mask and N95 mask filters. Additionally, the female talker’s ratings for naturalness, intelligibility, and listener effort were more affected than the male talker by the simple surgical and N95 mask filters.