Dec 17, 2020 · End-to-end (E2E) models have achieved promising results on multiple speech recognition benchmarks, and shown the potential to become the ...

End-to-end (E2E) models have achieved promising results on multi- ple speech recognition benchmarks, and shown the potential to be- come the mainstream. However ...

This paper focuses on incorporating contextual information into the continuous integrate-and-fire (CIF) based model that supports contextual biasing in a more ...

Recently, researchers have started leveraging the information of textual modality as additional contextual knowledge to help contextual ASR.

End-to-end (E2E) models have achieved promising results on multiple speech recognition benchmarks, and shown the potential to become the mainstream.

Collaborative decoding (ColDec) is proposed to customize the CIF-based ASR models. The application of ColDec in the field of ASR contextualization, ...

Mar 2, 2022 · Collaborative decoding (ColDec) [20] introduces phrase-level con- textual modeling and attention-based relevance modeling to con- textualize the ...

A PyTorch implementation of Continuous Integrate-and-Fire (CIF) module for end-to-end (E2E) automatic speech recognition (ASR).

Dec 17, 2020 · End-to-end (E2E) models have achieved promising results on multiple speech recognition benchmarks, and shown the potential to become the ...

This work presents a novel, all-neural, end-to-end (E2E) ASR system that utilizes such context, and jointly-optimizes the ASR components along with embeddings ...