BASENet: New Speech Enhancement Network Optimized for Human Hearing Achieves High Quality with Minimal Parameters
Researchers have developed BASENet, a speech enhancement model that allocates computational resources based on how human ears perceive different frequencies, using Bark-scale band partitioning and cross-band attention mechanisms. The architecture achieves state-of-the-art quality metrics (3.55 PESQ, 96% STOI) on standard benchmarks while using only 0.83 million parameters—fewer than competing methods with similar performance. This efficiency makes the model suitable for real-time speech enhancement on resource-constrained devices like mobile phones and embedded systems.
BASENet addresses a fundamental inefficiency in speech enhancement: most models apply uniform computational capacity across all frequencies despite human hearing having non-uniform spectral resolution. The proposed architecture partitions the frequency spectrum into Bark-scale bands—a perceptually-motivated division—and assigns scaled-capacity encoders to each band, automatically allocating deeper processing branches to perceptually dense low frequencies and lighter branches to high frequencies. A cross-band attention module captures harmonic dependencies across bands using frequency-pooled representations at linear computational complexity. Built on inverted residual blocks with dense connectivity and convolutional recurrent networks, BASENet achieves 3.55 PESQ and approximately 96% STOI on the VoiceBank+DEMAND benchmark with only 0.83 million parameters and 7.3 giga-multiply-accumulates—the fewest parameters among all methods achieving PESQ scores above 3.50. A causal variant (3.44 PESQ) designed for real-time processing surpasses several non-causal baselines, confirming suitability for streaming applications on resource-constrained devices.
What's missing
The paper does not discuss computational latency or inference time on actual mobile/embedded hardware, only theoretical MACs. Comparison with other perceptually-motivated speech enhancement approaches is limited. The study focuses on one benchmark (VoiceBank+DEMAND) and does not report generalization performance on other standard datasets or real-world noisy conditions beyond the benchmark's noise types.
What different sources said
- arXiv cs.AICenter
BASENet: Band-Adapted Speech Enhancement Network with Cross-Band Attention
Related
Topology-Aware Thermodynamics Improves DNA Probe Specificity Design
Researchers developed a new framework for designing DNA probes that accounts for the spatial organization of matched sequences, not just overall thermodynamic stability. Traditional methods rely on scalar measures like melting temperature and free energy, which miss how mismatches are distributed along the probe. The approach could improve diagnostic accuracy in applications like HPV detection and gene expression profiling.
Study Identifies Optimal Thermal Dose for Combining Focused Ultrasound with Immunotherapy in Tumors
Researchers used multimodal PET imaging to identify an optimal thermal dose range for focused ultrasound ablation that destroys tumor tissue while preserving conditions for immunotherapy delivery. The study found that excessive heating collapses blood vessels needed for antibody access, while insufficient heating fails to adequately reduce tumor burden. The findings could guide clinical design of combination treatments pairing thermal ablation with immunotherapies.
Plant MSH1 Protein Functions as Mismatch-Directed Nuclease for Organelle Genome Maintenance
Researchers have identified the precise mechanism by which the AtMSH1 protein in Arabidopsis plants recognizes and cleaves DNA mismatches and lesions, preventing mutations in organellar genomes. The protein combines a DNA mismatch recognition module with a nuclease domain that makes staggered cuts at specific positions relative to DNA damage. This discovery explains how plants maintain unusually low mutation rates in their mitochondrial and chloroplast DNA compared to other eukaryotes.