Perceptual linear predictive noise modelling for sinusoid-plus-noise audio coding

Abstract

Sinusoidal coding of audio subject to a bit-rate constraint will in general result in a noise-like residual signal. This residual signal is of high perceptual importance; reconstruction of audio using the sinusoidal representation only will typically result in an artificial sounding reconstruction. In this paper we present a method, called perceptual linear predictive coding (PLPC), where the residual is encoded by applying LPC in the perceptual domain. This method minimizes a perceptual modelling error and therefore represents only residual components that are of perceptual relevance, while automatically discarding components masked by the sinusoidal coded part. Subjective listening tests show that PLPC performs significantly better than ordinary LPC as a sinusoidal residual coding technique. Furthermore, PLPC combined with a flexible segmentation and model order allocation algorithm leads to a significant gain in terms of R/D performance for fragments with fast changing characteristics.

A Low-Complexity Spectro-Temporal Distortion Measure for Audio Processing Applications

“…This assumption is typically valid for quantization noise in audio coders but also in data-hiding applications like watermarking. The within-channel temporal envelope of can be expressed as (7) As a consequence of the averaging properties of the smoothing low-pass filter and the assumption that and are uncorrelated, it holds that (8) Motivated by this the following approximation is used: (9) By combining (9) and (6) we get (10) Next, we assume that only small errors are introduced to the clean signal which is typically the case in masking situations. Therefore, a good approximation of each element in the summation of (10) can be obtained by only taking into account the first term of the Maclaurin series expansion of .…”

“…In addition, the Par-model is defined as a mathematical norm, which allows for incorporating perceptual properties in least squares optimization algorithms. Examples are found in sinusoidal coding [8] and residual noise modeling [9]. Note that in the field of speech processing, mathematical tractable distortion measures are also used, like the log-spectral distance or distortion measures based on linear prediction (see, e.g., [10] and [11] for an overview).…”

Multichannel Audio Coding for Multimedia Services in Intelligent Environments

“…Regarding the noise part, it was not parametrically modeled for best audio quality. The work in [36] and more recently [37] has focused on the noise part modeling. In the former approach, the noise energy at each critical band was only retained, forming a perceptual spectral envelope of the noise signal.…”

A Multichannel Sinusoidal Model Applied to Spot Microphone Signals for Immersive Audio

“…In the first SNM derivation for audio signals [24], the noise part was modeled based on a piecewise-linear approximation of its short-time spectral envelope, or alternatively its linear predictive coding (LPC) envelope (assuming white noise excitation during synthesis). Popular methods for modeling the noise part have been described in [35] and [36]. In the former approach, the spectrum of the noise signal is divided into critical bands and the spectral envelope is estimated by retaining the energy in each band.…”