Alibaba open-sources two speech enhancement models: 1.5-2.3x faster separation, 22ms echo cancellation
aigclink · x · 2026-09-10
Alibaba open-sourced two new AI speech enhancement models, completing a suite of seven covering background noise, echo, and overlapping speakers. FLASepformer handles speech separation with linear attention, cutting quadratic compute to linear—1.5-2.3x faster with only 16-32% of the original memory. JAEC 16K is an interpretable echo canceller whose internal steps (delay estimation, echo estimation and suppression) are observable for debugging, with just 22ms latency for real-time calls; it doesn't yet handle nonlinear distortion like speaker clipping. Weights for both are released.
More from Models
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27