Speech intelligibility, related to the numbers of speech items recognized correctly by listeners, has received remarkable attention for several decades owing to its important role in speech communication and the evaluation for speech systems. How information pertaining to speech intelligibility is distributed among the various frequency bands of speech spectrum is always an arresting question in many applications.
Most of the preceding research investigated the effect of different frequency bands on the English intelligibility, including syllables, words and sentences. However, different languages are characterized by various specific acoustic and phonetic features. Besides, it is unclear that whether the primary findings about western languages are applicable to other languages, such as Mandarin Chinese. What’s more, related researches previously focused on the intelligibility of Chinese syllables but not sentences. However, intelligibilities will perform differently when speech materials change, and measures based on sentences are closer to the common communication scenarios in our daily life.
Recently, researchers CHEN Yafan, XU Yong and YANG Jun from the Institute of Acoustics of Chinese Academy of Sciences have investigated the relative spectral contributions of Chinese Mandarin sentences. Their research results provide valuable information and understandings for the perception of Chinese and English, especially for the further study of the Chinese speech intelligibility.
A Mandarin sentence intelligibility subjective test was implemented in a sound-proof room in this investigation. Chinese “everyday” sentences were low-pass, high-pass or bandpass filtered and restricted to different frequency regions. Twelve participants who were native speakers of Mandarin listened to stimuli and were required to write down each sentence they heard. The intelligibility score of each filtering condition for every listener was calculated by the percentage of key words written correctly for each list.
Listening experiments demonstrated that Different filtering conditions yielded significant effects on the intelligibility of Chinese sentences. Comparisons of the Mandarin sentence and KXY monosyllable intelligibility revealed that the low and middle frequency regions (<2100 Hz) made greater contribution to the intelligibility of Mandarin sentences than monosyllables. It provided advantages of lexical and semantic information on understanding Mandarin sentences.
Results about Chinese and English sentences demonstrated that the intelligibility redundancy on the frequency domain for Chinese was much lower than the redundancy for English. In addition, it could be concluded that the speaker’s first two formants were important to the sentence intelligibility for no matter English or Chinese.
Different materials with various phonemic features on frequencies affect the speech intelligibility. From this research, effect of various bands in the spectrum has been obtained preliminarily. Furthermore, how listeners combine information across frequency regions is important for understanding Chinese perception, which was not taken into consideration in the experiment. It is possible to further research the intelligibility of band-stop filtered Chinese speech for future.
Reference:
CHEN Yafan, XU Yong and YANG Jun. Intelligibilities of Filtered Chinese Mandarin Sentences. The 13th IEEE International Conference on Signal Processing (November 6th to 9th, 2016, Chengdu, China).
Contact:
CHEN Yafan
Key Laboratory of Noise and Vibration Research, Institute of Acoustics of Chinese Academy of Sciences, 100190 Beijing, China
Email: youryiyouxiang@126.com