eduzhai > Applied Sciences > Engineering >

Transfer Learning Approaches for Streaming End-to-End Speech Recognition System

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Transfer learning (TL) is widely used in conventional hybrid automatic speechrecognition (ASR) system, to transfer the knowledge from source to targetlanguage. TL can be applied to end-to-end (E2E) ASR system such as recurrentneural network transducer (RNN-T) models, by initializing the encoder and orprediction network of the target language with the pre-trained models fromsource language. In the hybrid ASR system, transfer learning is typically doneby initializing the target language acoustic model (AM) with source languageAM. Several transfer learning strategies exist in the case of the RNN-Tframework, depending upon the choice of the initialization model for encoderand prediction networks. This paper presents a comparative study of fourdifferent TL methods for RNN-T framework. We show 17 relative word error ratereduction with different TL methods over randomly initialized RNN-T model. Wealso study the impact of TL with varying amount of training data ranging from50 hours to 1000 hours and show the efficacy of TL for languages with smallamount of training data.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×