It Takes Few to TANGO: A Quantized Distributed Model for Binaural Speech Enhancement

2026-07-09

AI Summary

This work addresses the challenge of deploying multichannel speech enhancement systems on resource-constrained devices due to their high computational and memory demands. Focusing on the TANGO system, the authors propose MN-TANGO, a lightweight architecture that leverages the inherent robustness of spatial filtering to quantization errors to compensate for accuracy degradation in neural mask estimation, thereby simplifying the overall structure. The design incorporates INT8 weight and activation quantization, ERB-scale frequency band compression, and grouped recurrent layers to substantially reduce complexity. The resulting model achieves performance comparable to the original TANGO system while requiring only 4.65 MMAC/s of computation and 0.177 MB of memory.

Abstract

Neural network-based multichannel speech enhancement systems achieve strong enhancement performance, but their computational and memory requirements limit deployment on resource-constrained devices. This paper investigates low-precision inference for TANGO, a hybrid distributed binaural speech enhancement system combining neural mask estimation with spatial filtering. We evaluate post-training quantization and quantization-aware training for the neural components, and analyze how quantization errors in the mask estimators propagate through the downstream spatial filtering stage. Our analysis shows that, although quantization degrades intermediate mask estimates, the spatial filtering stage compensates for most quantization-induced errors. Leveraging this robustness, we simplify TANGO into MN-TANGO, reducing both model size and computational complexity while maintaining comparable final performance. By combining INT8 weight-and-activation quantization with ERB compression and grouped recurrent layers, the most compact MN-TANGO reaches 4.65 MMAC/s and 0.177 MB.

Problem

• binaural speech enhancement
• model quantization
• resource-constrained devices
• computational efficiency
• memory requirements

Innovation

• quantization
• binaural speech enhancement
• distributed model
• spatial filtering
• low-precision inference