Matxa-TTS v2 Central Catalan
Matxa-TTS v2 Central Catalan is BSC's grapheme-based speech model built with the Matcha-TTS flow-matching approach.
Model and training
It first trains on 36.2 hours of accent-balanced Catalan Common Voice material, then fine-tunes on OpenSLR69 and Festcat. The second stage freezes early text-encoder components while adapting other parts. Speech generation uses a separate vocoder, with AlVoCat documented in evaluation. Its GPL 3.0 terms differ from the multi-accent release's custom license.Matxa-TTS v2 Central Catalan: official model card
Practical use
Practical guidance. Use this model when the target is Central Catalan speech and a grapheme input pipeline is useful. Keep the acoustic model and vocoder versions together in an experiment record. Voice quality depends on both components, so changing the vocoder is not a neutral comparison of the same deployed system.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | matxa-tts-v2-ca-central-graphemes |
| License | GNU General Public License 3.0.Matxa-TTS v2 Central Catalan: official model card |
Related models
Matxa-TTS v2 Multi-Accent Catalan generates speech from graphemes with multiple Catalan accents and speakers.
Launch date not confirmedDownloadable- StyleTTS2 Catalan Multispeaker
Barcelona Supercomputing Center · BSC StyleTTS2
BSC's Catalan StyleTTS2 release trains a multispeaker speech synthesiser for Catalan.
Launch date not confirmedDownloadable