Frequency-domain bandwidth extension for low-delay audio coding applications
Stanislaw Gorlow · Common Library Network (Der Gemeinsame Bibliotheksverbund) · 2014
MPEG-4 Spectral Band Replication (SBR) is a sophisticated high-frequency reconstruction (HFR) tool for speech and natural audio which when used in conjunction with an audio codec delivers a broadband high-quality signal at a bit rate of 48 kbps or even below. The major drawback of this technique is that it significantly increases the delay of the underlying core codec. The idea of synthetic signal reconstruction is of particular interest also in real-time communications. There, a HFR method can be employed to further loosen the channel capacity requirements. In this thesis a delay-optimized derivative of SBR is elaborated, which can be used together with a low-delay speech and audio coder like the Fraunhofer ULD. The presented approach is based on a short-time subband representation of an acoustic signal of natural or artificial origin, and as such it utilizes a filter bank for the extraction and the manipulation of sound characteristics. The system delay for a combination of the ULD coder with the proposed low-delay bandwidth extension (LD-BWE) tool adds up to 12 ms at a sampling rate of 48 kHz. At the present stage, LD-BWE generates a subjectively confirmed excellent-quality highband replica at a simulated mean data rate of 12.8 kbps.