New approach to voiced onset detection in speech signal and its application for frame error concealment

This paper presents a new approach for accurate detection of voiced onsets in a speech signal. The knowledge of the precise location of the beginning of a voiced segment has several potential applications in speech compression (e.g. source-controlled variable bitrate coding, classification-based frame erasure concealment). The proposed onset detection algorithm is based on Teager's energy operator applied on sub-bands. The technique has been integrated in the signal classifier of the VMR-WB speech coding standard used for frame erasure (FE) concealment. A subjective listening test using MUSHRA methodology showed an improvement in frame error conditions with respect to the legacy VMR-WB classification.