公开(公告)号：US20240331685A1

公开(公告)日：2024-10-03

申请号：US18129996

申请日：2023-04-03

申请人： Deepgram, Inc.

发明人： Andrew Nathan Seagraves ,  Deepak Subburam ,  Adam Joseph Sypniewski ,  Scott Ivan Stephenson ,  Jacob Edward Cutter ,  Michael Joseph Sypniewski ,  Daniel Lewis Shafer

IPC分类号： G10L15/16 ,  G10L15/02 ,  G10L15/06

CPC分类号： G10L15/16 ,  G10L15/02 ,  G10L15/063 ,  G10L2015/025 ,  G10L2015/0633

摘要： An end-to-end automatic speech recognition (ASR) system can be constructed by fusing a first ASR model with a transformer. The input of the transformer is a learned layer generated by the first ASR model. The fused ASR model and transformer can be treated as a single end-to-end model and trained as a single model. In some embodiments, the end-to-end speech recognition system can be trained using a teacher-student training technique by selectively truncating portions of the first ASR model and/or the transformer components and selectively freezing various layers during the training passes.