About URGENT
URGENT is a series of speech enhancement challenges held at ICASSP 2026, Interspeech 2025 and NeurIPS 2024. The task is to develop a single model that handles several types of speech distortion and works at different sampling rates.
Systems are evaluated on simulated audio and real recordings using objective metrics and listening tests. The 2026 edition focuses on data selection and adds a speech quality assessment track. The 2025 edition introduced multilingual speech and a second track for larger training sets. The 2024 edition established the universal speech enhancement task.
Challenge series
Selected challenge
Participant and submission totals from the original challenge archive, including all evaluation phases.
Data & baselines
Evaluation datasets
Baseline models
Training corpora & licenses
The preparation scripts specify which subsets to use and how to process them.
Task rules & evaluation
Results
All results (Excel)Scroll for all metrics
No matching participants. Try a different name.
No participants selected. Select a name to show its results.
Farther out is better. Select a participant to hide or show.
Metrics & ranking
↑ Higher is better; ↓ lower is better. Parentheses show per-metric ranks. Scores and rankings are preserved as published. Select a column heading to sort, or restore the original order.
Radar charts show the top three entries by default. Select participant names to show more of the top 10. Each metric is scaled from 0 to 100% within the top 10, reversing lower-is-better metrics. As in the original charts, equal values are placed at 75% and missing scores at 0%. Hiding a participant does not change the scale.
Historical timeline
Dates are in Anywhere on Earth (UTC−12). All phases are closed.
Organizers
Affiliations at the time of the challenge.