book

Parametric Time-Frequency Domain Spatial Audio

by Ville Pulkki, Symeon Delikaris-Manias, Archontis Politis

December 2017

Intermediate to advanced

409 pages

14h 8m

English

Wiley-IEEE Press

Read now

Unlock full access

Notes
1.1 Introduction1.2 Time–Frequency Processing1.3 Processing of Spatial AudioNoteReferences
2.1 Introduction2.2 Sound Field Measurement by a Spherical Array2.3 Array Processing and Plane-Wave Decomposition2.4 Sensitivity to Noise and Standard Regularization Methods2.5 Optimal Noise-Robust Design2.6 Spatial Aliasing and High Frequency Performance Limit2.7 High Frequency Bandwidth Extension by Aliasing Cancellation2.8 High Performance Broadband PWD Example2.9 Summary2.10 AcknowledgmentReferences
3.1 Introduction3.2 The Plane-Wave Decomposition Problem3.3 Bayesian Approach to Plane-Wave Decomposition3.4 Calculating the IRLS Noise-Power Regularization Parameter3.5 Numerical Simulations3.6 Experiment: Echoic Sound Scene Analysis3.7 ConclusionsAppendixReferences
4.1 Introduction4.2 Parametric Processing OverviewReferences
5.1 Representing Spatial Sound with First-Order B-Format Signals5.2 Some Notes on the Evolution of the Technique5.3 DirAC with Ideal B-Format Signals5.4 Analysis of Directional Parameters with Real Microphone Setups5.5 First-Order DirAC with Monophonic Audio Transmission5.6 First-Order DirAC with Multichannel Audio Transmission5.7 DirAC Synthesis for Headphones and for Hearing Aids5.8 Optimizing the Time–Frequency Resolution of DirAC for Critical Signals5.9 Example Implementation5.10 SummaryReferences

6.1 Introduction6.2 Sound Field Model6.3 Energetic Analysis and Estimation of Parameters6.4 Synthesis of Target Setup Signals6.5 Subjective Evaluation6.6 ConclusionsNoteReferences
7.1 Introduction7.2 Parametric Sound Acquisition and Processing7.3 Multi-Wave Sound Field and Signal Model7.4 Direct and Diffuse Signal Estimation7.5 Parameter Estimation7.6 Application to Spatial Sound Reproduction7.7 SummaryNotesReferences
8.1 Introduction8.2 Notation and Signal Model8.3 Overview of the Method8.4 Loudspeaker-Based Spatial Sound Reproduction8.5 Binaural-Based Spatial Sound Reproduction8.6 ConclusionsReferences
9.1 Introduction9.2 Spectrogram Factorization9.3 Array Signal Processing and Spectrogram Factorization9.4 Applications of Spectrogram Factorization in Spatial Audio9.5 Discussion9.6 Matlab ExampleNoteReferences
10.1 Introduction10.2 Signal-Independent Enhancement10.3 Signal-Dependent EnhancementReferences
11.1 Introduction11.2 Notation and Signal Model11.3 Estimation of the Cross-Spectrum-Based Post-Filter11.4 Implementation Examples11.5 Conclusions and Further Remarks11.6 Source CodeNoteReferences
12.1 Introduction12.2 Time–Frequency Masks for Speech Enhancement Using Supervised Learning12.3 Artificial Neural Networks12.4 Mask Learning: A Simulated Example12.5 Mask Learning: A Real-World Example12.6 Conclusions12.7 Source CodeNotesReferences
13.1 Introduction13.2 Stereo-to-Multichannel Upmix Processor13.3 Digitally Enhanced Shotgun Microphone13.4 Surround Microphone System Based on Two Microphone Elements13.5 SummaryReferences
14.1 Introduction14.2 Parametric Sound Scene Synthesis for Virtual Reality14.3 Spatial Manipulation of Sound Scenes14.4 SummaryReferences
15.1 Introduction and Motivation15.2 Background15.3 Immersive Audio Communication System (ImmACS)15.4 Capture and Reproduction of Crowded Acoustic Environments15.5 ConclusionsNotesReferences

Content preview from Parametric Time-Frequency Domain Spatial Audio

6 Higher-Order Directional Audio Coding

Archontis Politis and Ville Pulkki

Department of Signal Processing and Acoustics, Aalto University, Finland

6.1 Introduction

The first-order model of directional audio coding (FO-DirAC), detailed in Chapter 5, estimates a single direction of arrival (DoA) and a global diffuseness estimate that depend on the interactions of all the sound wave contributions arriving in a single observation window. If two source signals with similar spectra and incident from different directions are captured with nearly equal powers at the same time–frequency point, three scenarios can occur during FO-DirAC reproduction with perceivable effects. If the analysis of the B-format signals is computed through very short time averages, then the estimated DoA will fluctuate between the two true DoAs, with most estimates clustered around the true values. If no temporal smoothing is performed on the respective time–frequency gains applied during synthesis, then this directional distribution will be reproduced appropriately, with most power spatialized around the two DoAs, but with a high chance of perceived image instability and musical noise. If longer temporal smoothing is used on the spatialization gains, then most likely multiple loudspeakers will be activated between the two DoAs, resulting in a stable rendering but with an increased spatial blurring similar to linear or ambisonic decoding. On the other hand, if the DoA and diffuseness are computed using longer ...