Speaker: William Fernando López Gavilánez.
Abstract: Automatic speech recognition has advanced rapidly in recent years, yet fine-grained evaluation for Iberian languages remains limited. In this work, we benchmark several open-weights speech-to-text models across multiple Iberian languages (Basque, Catalan, Galician, Portuguese, and Spanish) as well as additional languages such as Turkish and German. We conduct a systematic comparison of recognition performance and computational efficiency, reporting word error rate, real-time factor, and inverse real-time factor. These metrics enable a fair accuracy and efficiency analysis, providing practical guidance for selecting open ASR models in multilingual and resource-constrained scenarios.