Whisper Balearic

BSC's Balearic Whisper release adapts OpenAI Whisper Large v3 to Balearic-accented Catalan.

Model and training

The model was fine-tuned on 90 hours from the perfect_matches split of 3catparla_asr. The source consists of manually transcribed television speech. It produces unpunctuated text and has an official faster-whisper conversion. Its regional training contribution makes it a separate checkpoint from the Valencian and general Catalan models.Whisper Balearic: official model card

Practical use

Practical guidance. Compare it on Balearic speakers whose recording conditions differ from television material. Accent adaptation does not ensure equal results across every island, age group or speaking style. Inspect local words and short function words, and check whether the model improves recognition without changing the intended wording.

Specifications & access

Whisper Balearic specifications
DetailValue and conditions
AccessDownloadable
Model identifierswhisper-3cat-balearicfaster-whisper-3cat-balearic
LicenseApache License 2.0.Whisper Balearic: official model card

Related models