Granite Speech 5.0 470M TurboCTC NC
TurboCTC NC is a separate 470M English speech recognition checkpoint. It uses a non-commercial, share-alike license. The Apache license of the main multilingual TurboCTC release does not apply to these weights.
Research release
Model design and capabilities
The model uses a 16-block Conformer CTC encoder and a 16,384-unit BPE output vocabulary. It is available for research under CC BY-NC-SA 4.0. Its language coverage and training data differ from the main multilingual release.Granite Speech 5.0 470M TurboCTC NC model card
Practical use
Practical guidance. Use this checkpoint only where its non-commercial terms fit the project. Keep its evaluation results and downloaded files distinct from the Apache release. Otherwise a successful experiment can be incorrectly treated as evidence for a deployable commercial configuration.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | ibm-granite/granite-speech-5.0-470m-turboctc-nc |
| License | CC BY-NC-SA 4.0Granite Speech 5.0 470M TurboCTC NC model card |
| Language | EnglishGranite Speech 5.0 470M TurboCTC NC model card |
| Training audio | Approximately 75,000 hours from public corpora.Granite Speech 5.0 470M TurboCTC NC model card |
Related models
- Granite Speech 5.0 470M TurboCTC
IBM ยท Granite Speech
A non-autoregressive speech recognizer that produces capitalized, punctuated text in seven languages.
Launch date not confirmedDownloadable