Speech intelligibility of English, Polish, Arabic and Mandarin under different room acoustic conditions

Oral communication is at the center of multilingual and multicultural interactions among people in public, social and commercial spaces. However, only a few studies have been comparing the differences between physical measures and subjective measures of speech intelligibility for native speakers of different languages, and most studies have concentrated on comparisons between English and Chinese. Furthermore, speech intelligibility design guidelines typically focus on physical parameters only, such as the speech transmission index, signal-to-noise ratio and reverberation time, disregarding the possibility of having interactions between room acoustic parameters and the languages spoken.

Laurent Galbrun and Kivanc Kitapci from Heriot-Watt University, United Kingdom, examined the speech intelligibility of English, Polish, Arabic and Mandarin under different room acoustic conditions. Their study aimed to identify speech intelligibility variations between languages, the extent of these variations and their statistical significance. The work is now published in the peer-reviewed journal, Applied Acoustics.

The English, Polish, Arabic and Mandarin languages were selected for being representative of a wide range of linguistic properties. In order to assess the intelligibility of each language, diagnostic rhyme tests, phonemically balanced word lists and phonemically balanced sentence lists were employed.
All the intelligibility tests were carried out at four different room acoustic conditions, which were described in terms of different speech transmission index values. According to speech transmission index qualification ratings, the values tested were rated as “bad”, “poor”, “good”, and “excellent” speech intelligibility conditions.

Speech intelligibility material was recorded using six speakers per language (three males and three females). Owing to the variety of accents within languages, attention was given to the origin of the speakers. For instance, English speakers had to speak English with Received Pronunciation that is normally associated with formal speech.

The listening tests included six participants per language (three males and three females). These tests were undertaken in a chamber whose sound absorption properties could be controlled by adding or removing panel absorbers. Furthermore, white noise was used to achieve different signal-to-noise ratios. The combination of these different sound absorption and signal-to-noise ratio settings allowed achieving the four different room acoustic conditions tested.

The word intelligibility results revealed that statistically significant differences occurred between languages across most of the speech transmission index conditions tested. Word scores indicated that English was the most intelligible language, analysis suggesting that this may be attributed to its low consonant-to-vowel ratio, its larger high frequency content, as well as its larger temporal variability and dynamic range at high frequencies. Mandarin’s speech intelligibility was also good across most conditions, as Mandarin could take advantage of an average consonant-to-vowel ratio, a fairly high temporal variability at high frequencies, and tonality. On the other hand, Arabic and Polish were found to be the most sensitive languages to artificial noise and increased reverberation time, which resulted in poor speech intelligibility. This was attributed to their moderately high and high consonant-to-vowel ratios respectively, as well as low high frequency content and temporal variations. Sentence intelligibility scores confirmed variations between languages, but these were statistically significant only under the “poor” room acoustic condition (sentence tests being less sensitive to very god and very poor room acoustic conditions). Overall, the results of the study revealed that each language is affected differently by room acoustic properties, and that variations can be significant and are due to the differences between the linguistic and phonological properties of each language. The findings indicated that recommendations solely based on room acoustic parameters might then prove to be insufficient for designing multilingual environments.