Zero-bit self-adaptive spectral noise shaping for speech and audio coding.

Area of Science:

• Signal Processing
• Audio Engineering
• Information Theory

Background:

• Quantization noise in audio coding is often modeled as white noise, causing audible distortions.
• Noise shaping techniques redistribute noise to less perceptible frequencies, improving perceived audio quality.
• Existing methods for adaptive noise shaping involve forward or backward coefficient estimation, which can introduce lag or require additional bitrate.

Purpose of the Study:

• To introduce a novel, bitrate-efficient method for perceptually shaping coding noise in speech and audio codecs.
• To develop a new approach for estimating adaptive linear filter coefficients for noise shaping.
• To implement and evaluate a "zero-bit" constrained adaptation technique.

Main Methods:

• The proposed method utilizes pre- and post-processing adaptive linear filters.
• Filter coefficients are estimated directly from the current input signals for both pre- and post-processors.
• A "constrained adaptation" approach is implemented, imposing specific constraints on coefficient calculation.

Main Results:

• The "zero-bit" constrained adaptation method effectively shapes coding noise.
• Performance is comparable to traditional forward and backward adaptation techniques.
• The method introduces no additional bitrate cost or temporal lag.

Conclusions:

• Constrained adaptation offers an efficient and effective solution for noise shaping in audio coding.
• This bitrate-free, lag-free approach enhances perceived audio quality.
• The technique holds promise for improving speech and audio codec performance.