Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Audio dequantization using instantaneous frequency

Vojtěch Kovanda, Pavel Rajmic

TL;DR

The paper tackles audio dequantization at low bit depths by introducing PHADQ, a phase-aware regularization framework that preserves energy and sinusoidal continuity in a phase-corrected time-frequency representation. Building on PHAIN, it formulates a convex optimization with a penalty $\\lambda \\| D R_{\\omega_s} G_g x \\|_1$ and a quantization-consistent feasible set $\\Gamma$, and proposes two variants, B-PHADQ and U-PHADQ, differing in how the instantaneous frequency $\\omega_s$ is handled. Experimental results on IRMAS and EBU SQAM show that PHADQ can achieve perceptual improvements (ODG) over a sparsity-based baseline while requiring fewer iterations, though SDR may favor the baseline in some cases. The work highlights the practical potential of phase-aware dequantization for efficient, high-quality restoration of quantized audio, with open-source Matlab code for replication.

Abstract

We present a dequantization method that employs a phase-aware regularizer, originally successfully applied in an audio inpainting problem. The method promotes a temporal continuity of sinusoidal components in time-frequency representation of the audio signal, and avoids energy loss artifacts commonly encountered with l1-based regularization approaches. The proposed method is called the Phase-Aware Audio Dequantizer (PHADQ). The method are evaluated against the state-of-the-art using the SDR and PEMO-Q ODG objective metrics, and a~subjective MUSHRA-like test.

Audio dequantization using instantaneous frequency

TL;DR

The paper tackles audio dequantization at low bit depths by introducing PHADQ, a phase-aware regularization framework that preserves energy and sinusoidal continuity in a phase-corrected time-frequency representation. Building on PHAIN, it formulates a convex optimization with a penalty and a quantization-consistent feasible set , and proposes two variants, B-PHADQ and U-PHADQ, differing in how the instantaneous frequency is handled. Experimental results on IRMAS and EBU SQAM show that PHADQ can achieve perceptual improvements (ODG) over a sparsity-based baseline while requiring fewer iterations, though SDR may favor the baseline in some cases. The work highlights the practical potential of phase-aware dequantization for efficient, high-quality restoration of quantized audio, with open-source Matlab code for replication.

Abstract

We present a dequantization method that employs a phase-aware regularizer, originally successfully applied in an audio inpainting problem. The method promotes a temporal continuity of sinusoidal components in time-frequency representation of the audio signal, and avoids energy loss artifacts commonly encountered with l1-based regularization approaches. The proposed method is called the Phase-Aware Audio Dequantizer (PHADQ). The method are evaluated against the state-of-the-art using the SDR and PEMO-Q ODG objective metrics, and a~subjective MUSHRA-like test.
Paper Structure (10 sections, 6 equations, 4 figures, 1 table, 1 algorithm)

This paper contains 10 sections, 6 equations, 4 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Review of PHAIN (Phase-aware audio inpainter)
  3. Method
  4. B-PHADQ
  5. U-PHADQ
  6. Experiment and Results
  7. Objective evaluation
  8. Subjective evaluation
  9. Computational considerations
  10. Conclusion

Figures (4)

  • Figure 1: Reconstruction quality of a violin recording (6 bps word length) across iterations in terms of SDR. For U-PHADQ, $\omega_{\mathbf{s}}$ is updated every 10 iterations.
  • Figure 2: Average SDR and ODG on the EBU SQAM dataset
  • Figure 3: Average SDR and ODG results on the IRMAS dataset
  • Figure 4: MUSHRA results for 6 and 7 bps quantization on EBU SQAM dataset