Catalan ASR Arena

Performance Analysis

Model Analysis

Explore how Catalan speech recognition models balance transcription accuracy and inference speed, then compare the strongest results for each benchmark metric.

Accuracy vs. Speed

Pareto Frontier

Models on the frontier offer the best available trade-offs between weighted average WER and RTFx.

All models Pareto frontier

Top 10

Weighted Average WER by Model

Top 10

Weighted Average CER by Model

Top 10

Weighted Average RTF by Model

Model Submission

Submit your model to the leaderboard

Share the basic details of your Hugging Face model and we will review it for evaluation on the Catalan ASR benchmark.

Before you submit Make sure the model is public on Hugging Face and that the model card includes the main information about the system.

Sign In

Hugging Face account

Sign in with HF

Sign in with Hugging Face to enable model submission.

Submission Form

Tell us about your model

You must be logged in with Hugging Face to submit this form.

Hugging Face model id Use the full repository name exactly as it appears on Hugging Face. Base model Required. Use the full base model repo URL or identifier. Architecture Model size

Allowed range: 1MB to 500GB. Number of parameters

Use billions or millions. Allowed range: 0M to 500B. Model type Languages The current benchmark only accepts Catalan submissions. License

About

Catalan ASR Leaderboard

This leaderboard helps users discover and compare speech recognition models for Catalan. It brings together benchmark results in a clear public view so researchers, developers, and organizations can follow progress over time.

Contributors

Built with support from BSC and AINA

Barcelona Supercomputing Center

BSC is Spain's national supercomputing center and contributes research and infrastructure that support language technology and evaluation.

Visit BSC

AINA

AINA is a Catalan language technology project that promotes open resources and tools to strengthen digital use of Catalan.

Visit AINA