Nvidia patents method to combine language tools for multilingual speech recognition
Nvidia describes a method for creating a single, combined tool that handles multiple languages by grouping similar ones together. This hybrid tool is then used to train a model that converts spoken words into written text. The approach aims to improve how speech recognition systems process different languages simultaneously.
In the filing’s own words
Disclosed are apparatuses, systems, and techniques that implement training and deployment of hybrid tokenizers for use in speech and/or text processing models. The techniques include grouping monolingual tokenizers (MTs) into clusters that include one or more MTs selected based at least on linguistic similarity of the one or more MTs. An individual MT includes at least a plurality of language units of a respective language. The techniques include merging the MT(s) of individual clusters to obtain respective cluster tokenizers. Obtaining a first cluster tokenizer includes combining linguistic information associated with an individual token of two or more MTs of a first cluster. The techniques further include combining the plurality of cluster tokenizers into a hybrid tokenizer and training, using the hybrid tokenizer, a speech-to-text model to generate a text associated with a speech input into the model.
What exactly they claim (claim 1)
1. A method comprising: grouping a plurality of monolingual tokenizers (MTs) into a plurality of clusters, an individual cluster of the plurality of clusters including one or more MTs of the plurality of MTs selected based at least on linguistic similarity of the one or more MTs, an individual MT of the plurality of MTs including at least a plurality of language units of a respective language of a plurality of languages; merging the one or more MTs of individual clusters to obtain respective cluster tokenizers of a plurality of cluster tokenizers, obtaining a respective cluster tokenizer of the plurality of cluster tokenizers including combining linguistic information associated with an individual token of two or more MTs of a first cluster of the plurality of clusters; combining the plurality of cluster tokenizers into a hybrid tokenizer; and training, using the hybrid tokenizer, a speech-to-text (S2T) model to generate a text associated with a speech input into the S2T model.
Facts
- Company
- Nvidia (NVIDIA Corporation)
- Official title
- MULTILINGUAL AUTOMATIC SPEECH AND TEXT PROCESSING FOR CONVERSATIONAL AI SYSTEMS AND APPLICATIONS
- Publication number
- US20260268889A1
- Application number
- 19072831
- Filed
- Mar 6, 2025
- Published
- Sep 10, 2026
- Inventors
- 4
- Claims
- 20
Classified under
G10L 13/08 Physics & computingG10L 13/027 Physics & computingG10L 13/04 Physics & computingMore from Nvidia in AI & software
- Nvidia patents AI coach that analyzes your gameplay to suggest strategies Oct 13, 2026
- Nvidia patents AI system that writes and voices character conversations Sep 10, 2026
- Nvidia patents system that uses AI to find bugs in game videos Sep 17, 2026
- Nvidia patents AI method to make digital characters move realistically Sep 8, 2026
- Nvidia patents AI system that builds image generators from text prompts Sep 15, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.