Open-source VoxLock tops Krisp voice isolation benchmark with 4.95% WER
Arbaz Siddiqui · hf · 2026-11-02
- VoxLock is a target-speaker extraction model for voice agents that needs no speaker enrollment step, isolating the caller's voice from noise or a second speaker.
- On the Krisp Voice Isolation Benchmark (265 real recordings), VoxLock achieves 4.95% word error rate and 0.11 background words leaked per minute, beating paid rivals Krisp VI 2.5 (5.30%) and ai-coustics VF L (6.84%), versus 23.21% with no processing.
- The fine-tuned causal model reaches near-commercial speech recognition accuracy on real recordings with low CPU usage and latency, suited for real-time voice agent deployment.
More from Models
- Claude upgrade-page traffic rebounded to 17.6M in September, near May's 18.6M peak — firstadopter · 2026-11-02
- Open-source GLiNER-PII scrubs 13 identifiers from a discharge note in 550ms on GPU — MaziyarPanahi · 2026-11-02
- OpenAI Dots user says bloat and broken infra drove him to try newcomer Muse — Koala_Confused · 2026-11-02
- Fable 5.5 reportedly one-shots animation in pure code as Anthropic's coding bet pays off — Hesamation · 2026-11-02
- One agent goal has been running for 4 days, burning 30K tokens — CtrlAltDwayne · 2026-11-02
- Daily AI Brief: Gemini 4 Carbon spotted on Arena, DeepSeek open-sources coding agent — testingcatalog · 2026-11-02