Design And Implementation Of Parametric Codec For Audio Signals Sampled With High Frequencies

Parametric audio coding has became a more and more important supplement for traditional transform audio codec, especially in applications using low-bitrate audio coding, such as Digital Radio Mondiale, mobile communication and multimedia information search. The research on source models is one of the most crucial issues in parametric audio coding. A general review for the developments in parametric audio coding and the evolution of source models is presented in this paper. Brief introductions of some representative source models are also included in this paper.Many existing source models were designed to process audio signals sampled at low frequencies, such as 32KHz, 22KHz, 16KHz, even 8KHz. With the low sampling frequency, the final bitrate of the encoded files can be reduced significantly. But the limited bandwidth does deteriorate the synthesis sound quality simultaneously.With improvements of some current methods used in parametric audio coding, a parametric audio codec system is presented and implemented in this paper, which is oriented to high sampling frequency audio signals. The proposed system is composed of a multiresolution sinusoidal model and a multiresolution ERB noise model, both of which can make a fair trade-off between bitrate and sound quality. As options for the system, harmonic structure analysis, dual-transform transient component detector and dual-transform transient model also make the system perform better for certain kinds of audio signals. The user can determine how to use these optional modules to improve synthesis sound quality for special audio signals. With all these models' help, the proposed parametric audio coding system can achieve an average bitrate of 24Kbps/ch for audio signals sampled at 44.1KHz/48KHz with a good subjective quality.