A unified beamforming and source separation model for static and dynamic human-robot interaction

Jorge Wuth ¹, Rodrigo Mahu ¹, Israel Cohen ²

⁰Speech Processing and Transmission Laboratory, Department of Electrical Engineering, University of Chile, Av. Tupper 2007, Santiago, Chile.

Jasa Express Letters +

|

March 5, 2024

View abstract on PubMed

Summary

This study introduces a unified model for combining beamforming and blind source separation (BSS). The novel approach significantly improves signal-to-noise ratio (SNR) in challenging human-robot interaction (HRI) scenarios.

Area Of Science

Signal Processing
Acoustics
Robotics

Background

Combining beamforming and blind source separation (BSS) is crucial for enhancing speech intelligibility in noisy environments.
Existing methods often struggle in dynamic or complex acoustic settings like human-robot interaction (HRI).

Purpose Of The Study

To propose a unified model integrating beamforming and BSS for improved speech recovery.
To evaluate the model's performance in real-world static and dynamic HRI data.

Main Methods

Developed a unified model combining BSS with the minimum-variance distortionless response (MVDR) beamformer.
Validated model assumptions using Oracle information for accurate target speech recovery.
Tested the system on real static HRI data and analyzed performance in dynamic HRI environments.

Main Results

The unified model achieved higher signal-to-noise ratio (SNR) compared to previous parallel and cascade systems on static HRI data.
In dynamic HRI environments, the proposed system demonstrated a 2.8 dB greater SNR gain than cascade systems.
Parallel combinations were found to be infeasible in the dynamic HRI setting.

Conclusions

The unified model offers a superior approach for combining BSS and beamforming.
The method effectively enhances speech signal quality in challenging HRI scenarios.
This integration is particularly beneficial for dynamic and difficult-to-model acoustic environments.