Audio Signal Processing and Coding

4.6 LINEAR PREDICTION IN WIDEBAND CODING

Until now, we discussed the use of LP in narrowband coding with signal band-width limited to 150–3400 Hz. Signal bandwidth in wideband speech coding spans 50 Hz to 7 kHz; which substantially improves the quality of signal reconstruction, intelligibility, and naturalness. In particular, the introduction of the low-frequency components improves the naturalness, while the higher frequency extension provides more adequate speech intelligibility. In case of high-fidelity audio, it is typical to consider sampling rates of 44.1 kHz and signal bandwidth can range from 20 Hz to 20 kHz. Some of the recent super high-fidelity audio storage formats (Chapter 11) such as the DVD-audio and the super audio CD (SACD) consider signal bandwidths up to 100 kHz.

4.6.1 Wideband Speech Coding

Over the last few years, several wideband speech coding algorithms have been proposed [Orde91] [Jaya92] [Lafl93] [Adou95]. Some of the coding principles associated with these algorithms have been successfully integrated into several speech coding standards, for example, the ITU-T G.722 subband ADPCM standard and the ITU-T G.722.2 AMR-WB codec.

4.6.1.1 The ITU-T G.722 Codec

The ITU-T G.722 standard (Figure 4.9) uses a combination of both subband and ADPCM (SB-ADPCM) techniques [G.722] [Merm88] [Span94] [Pain00]. The input signal is sampled at 16 kHz and decomposed into two subbands of equal bandwidth using quadrature mirror filter (QMF) banks. The subband filters h_low(n) and ...