On improving speech intelligibility in automotive hands-free systems

摘要

Due to its narrow acoustic bandwidth, typical telephone speech is poorly intelligible on a phoneme or syllable basis. Artificial bandwidth extension (ABWE) techniques intend to increase speech intelligibility by estimating and reconstructing higher frequency components up to 7kHz. Hands-free systems in automotive environments could especially benefit from that, since drivers are mentally stressed during telephone conversations. By improving speech intelligibility, their mental load may decrease. Thus driving safety and comfort increase. This paper presents a real-time ABWE implementation integrated into an automotive hands-free system. Its novelty is a phoneme-specific design that yields an improved performance on critical phonemes, especially /s/ and /z/. Hence, typical lisping effects of previous ABWE approaches are reduced. We performed informal listening tests on meaningless vowel-consonant-vowel (VCV) syllables demonstrating a significantly improved fricative intelligibility of ABWE speech vs. NB speech.