Conformer Transducer Languages of Spain Punctuated
This BSC LoS Conformer release adds trained punctuation to multilingual speech recognition.
Model and training
It derives from NVIDIA's Spanish Conformer Transducer Large and uses about 2,700 hours of Catalan, Spanish, Galician and Basque audio. The roughly 120M-parameter checkpoint is distinct from the plain-output LoS model. The published workflow uses NeMo, so its runtime requirements differ from a Transformers Whisper deployment.Conformer Transducer Languages of Spain Punctuated: official model card
Practical use
Practical guidance. Assess whether punctuation reduces editing work without introducing misleading sentence boundaries. Compare the punctuated and plain versions on identical segmented audio. A smaller network can be operationally useful, but that alone does not establish lower end-to-end latency once preprocessing and batching are included.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | stt_los_conformer_transducer_large_punctuated |
| License | Apache License 2.0.Conformer Transducer Languages of Spain Punctuated: official model card |
| Model size | Approximately 120 million parameters.Conformer Transducer Languages of Spain Punctuated: official model card |
Related models
- Conformer Transducer Languages of Spain
Barcelona Supercomputing Center · BSC LoS Conformer
BSC's LoS Conformer adapts NVIDIA's Spanish Conformer Transducer to four languages of Spain.
Launch date not confirmedDownloadable This BSC Whisper adaptation produces punctuated transcripts in Spanish, Catalan, Galician and Basque.
Launch date not confirmedDownloadable