New 'Listen First' approach boosts hearing aid speech clarity by evaluating output
Traditional speech enhancement fails in noisy rooms; new output-based method fixes that.
Hearing-aid speech enhancement has traditionally depended on input-based feature estimation, often using a voice activity detector (VAD) to configure beamformers. But in challenging acoustic scenes—precisely when users need the most help—these features extracted from noisy microphone signals become unreliable, leading to poor performance. A team of researchers from Aalborg University and Oticon (Panos Apostolidis, Svend Feldt, Zheng-Hua Tan, Jan Østergaard, Jesper Jensen) propose a radical new paradigm: determine system settings by evaluating characteristics of the system's output. They call it 'Listen first.'
To demonstrate, they implement an output-based system that selects among a set of minimum power distortionless response (MPDR) beamformers—a class typically avoided due to sensitivity to steering errors. However, within the output-based framework, MPDR beamformers become effective and stable. In experiments comparing against a standard input-based minimum variance distortionless response (MVDR) baseline, the proposed system consistently outperforms it, with the largest gains at low signal-to-noise ratios. The improvement is confirmed across multiple metrics: SNR, extended short-time objective intelligibility (ESTOI), and perceptual evaluation of speech quality (PESQ). The paper has been accepted at the International Workshop on Acoustic Signal Enhancement (IWAENC) 2026.
- Traditional input-based VAD features become unreliable in noisy environments where hearing aid users need help most.
- New system selects among MPDR beamformers by evaluating output characteristics, turning a typically avoided technique into a powerful tool.
- Outperforms standard MVDR baseline at low SNRs across SNR, ESTOI, and PESQ metrics.
Why It Matters
Could dramatically improve hearing aid performance in real-world noisy environments, helping millions hear clearly.