Result Details
EU-ASR
Created: 2025
Type
software
Language
English
Authors
Szőke Igor, Ing., Ph.D., DCGM (FIT)
Veselý Karel, Ing., Ph.D., DCGM (FIT)
Polok Alexander, Ing., DCGM (FIT)
Salimbajevs Askars
Veselý Karel, Ing., Ph.D., DCGM (FIT)
Polok Alexander, Ing., DCGM (FIT)
Salimbajevs Askars
Description
EU-ASR codebase, includes code for data-formatting, self-supervised pre-training, fine-tuning training and deploying of a monolingual streaming ASR. It is the output of the EU Tender contract CNECT/LUX/2022/OP/0030 - LC-01940589 - Lot 2. The underlying technology is based on `pytorch, transformers, k2-fsa, sherpa-onnx`, offering SOTA Transducer architecture with E-Branchformer encoder. The runtime is a modular python-based client-server application with GRPC or websocket APIs. The runtime can process an uploaded file or a continuous stream of audio (streaming ASR).
Keywords
speech recognition, training code, deployment code
URL
License
The result is being used by the owner
License Fee
The licensor does not require a license fee for the result
Projects
Answer to EC Tender CNECT/LUX/2022/OP/0030 - LANGUAGE TECHNOLOGY SOLUTIONS Lot 2, EU, Digital Europe Programme, start: 2023-01-01, end: 2025-03-31, completed
Research groups
Speech Data Mining Research Group BUT Speech@FIT (RG SPEECH)
Departments