PERCEPTUAL RESTORATION OF INTERMITTENT SPEECH UNDER NOISY ENVIRONMENT

Speech communication is achieved through communication networks as well as face-to-face conversations. In digital communication, VoIP, which packetizes a speech signal into the standardized packets, is widely employed for achieving speech communication. It, is however, faced with a serious problem, that is, packet loss. The packet loss becomes much serious due to the global explosion of packet distribution. It means that investment in communication infrastructures could not solve this problem essentially. Intermittent speech should be restored on a user’s client equipment. As packet loss concealment on the client side, a waveform substitution is popular and standardized by ITU. The ITU-T recommendation G.711 conceals packet loss by inserting a phase-adjusted amplitude-attenuating previous packet into each break, but cannot deal with long-term breaks over 60 ms. The authors have previously proposed a novel approach to packet loss concealment relying on the human auditory capability. This method does not aim at restoring a waveform of an intermittent speech signal, but achieves perceptual restoration relying on the phonetic restoration effect. When the gaps of the intermittent speech signal are filled up with a loud signal, we can listen to the restored speech smoothly even if some segments of the original speech signal are completely lost. In this paper, the gap-filling signal is optimized under noisy environments. It is found that the suitable level of the gap-filling signal varies depending on signal-to-noise ratios. Performance evaluation for the proposed method is carried out by listening test under various simulated noise environments.