Subjective and Objective Quality Assessment for Noise Reduced Speech

Recent development in telecommunication services, such as VoIP in NGN and car communications has become increasingly necessary for use of hands-free communication system using separate microphones and loudspeakers. Hands-free system has been largely affected by noisy circumstances. This paper describes experimental results and perspectives for subjective and objective quality assessment of noise reduced speech. In this paper, noisy circumstances, such as subway, babble, car, and exhibition were considered, and as the noise reduced algorithms, spectral subtraction with smoothing of the time direction, temporal domain SVD-based speech enhancement, GMM-based speech estimation, and KLT-based comb-filtering were used. Noise reduced speech were subjectively assessed by word intelligibility and opinion rating methods. Objective quality assessment was studied from viewpoints of objective quality measure using perceptual evaluation of speech quality (PESQ) in the auditory domain. First, this paper describes subjective and objective quality assessment from viewpoints of the opinion rating methods, and then, word intelligibility.