ConMamba Small Catalan
ConMamba Small Catalan is BSC's Catalan speech recogniser built with convolution-augmented Mamba state-space layers.
Model and training
The model was trained on 4,929 hours of Catalan speech. It produces unpunctuated transcripts. Its acoustic architecture differs from BSC's Whisper adaptations; the card documents a ConMamba training release rather than an OpenAI checkpoint conversion. The weights use GPL 3.0 terms.ConMamba Small Catalan: official model card
Practical use
Practical guidance. This model offers an architecture comparison for Catalan recognition. Evaluate memory use and response time with the actual implementation rather than inferring them from the state-space design alone. Compare long recordings, background noise and regional accents with the Whisper baseline under the same audio preprocessing.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ConMamba-small-ca |
| License | GNU General Public License 3.0.ConMamba Small Catalan: official model card |
Related models
- ConMamba Small Spanish
Barcelona Supercomputing Center · BSC ConMamba
ConMamba Small Spanish is BSC's Spanish speech-recognition release using a convolution and state-space architecture.
Launch date not confirmedDownloadable BSC's Catalan Whisper model is a trained adaptation of OpenAI Whisper Large v3 for speech recognition.
Launch date not confirmedDownloadable