2021 IEEE International Conference on Acoustics, Speech and Signal Processing

6-11 June 2021 • Toronto, Ontario, Canada

Extracting Knowledge from Information

2021 IEEE International Conference on Acoustics, Speech and Signal Processing

6-11 June 2021 • Toronto, Ontario, Canada

Extracting Knowledge from Information

Technical Program

Paper Detail

Paper IDSPE-40.4
Paper Title EMFORMER: EFFICIENT MEMORY TRANSFORMER BASED ACOUSTIC MODEL FOR LOW LATENCY STREAMING SPEECH RECOGNITION
Authors Yangyang Shi, Yongqiang Wang, Chunyang Wu, Ching-Feng Yeh, Julian Chan, Frank Zhang, Duc Le, Mike Seltzer, Facebook AI, United States
SessionSPE-40: Speech Recognition 14: Acoustic Modeling 2
LocationGather.Town
Session Time:Thursday, 10 June, 15:30 - 16:15
Presentation Time:Thursday, 10 June, 15:30 - 16:15
Presentation Poster
Topic Speech Processing: [SPE-RECO] Acoustic Modeling for Automatic Speech Recognition
IEEE Xplore Open Preview  Click here to view in IEEE Xplore
Virtual Presentation  Click here to watch in the Virtual Conference
Abstract This paper proposes an efficient memory transformer Emformer for low latency streaming speech recognition. In Emformer, the long-range history context is distilled into an augmented memory bank to reduce self-attention’s computation complexity. A cache mechanism saves the computation for the key and value in self-attention for the left context. Emformer applies a parallelized block processing in training to support low latency models. We carry out experiments on benchmark LibriSpeech data. Under average latency of 960 ms, Emformer gets WER 2.50 % on test-clean and 5.62% on test-other.Comparing with a strong baseline augmented memory transformer(AM-TRF), Emformer gets 4.6 folds training speedup and 18% relative real-time factor (RTF) reduction in decoding with relative WER reduction 17% on test-clean and 9% on test-other. For a low latency scenario with an average latency of 80 ms, Emformer achieves WER 3.01% on test-clean and 7.09% on test-other. Comparing with the LSTM baseline with the same latency and model size, Emformer gets relative WER reduction 9% and 16% on test-clean and test-other, respectively.