1. [Publications](/publications)
2. Multi-blank Transducers for Speech Recognition
 
 # Multi-blank Transducers for Speech Recognition

  ![Publication image](/sites/default/files/styles/wide/public/default_images/default.jpeg?itok=qUFsuJCP "Publication image")

 This paper proposes a modification to RNN-Transducer (RNN-T) models for automatic speech recognition (ASR). In standard RNN-T, the emission of a blank symbol consumes exactly one input frame; in our proposed method, we introduce additional blank symbols, which consume two or more input frames when emitted. We refer to the added symbols as big blanks, and the method multi-blank RNN-T. For training multi-blank RNN-Ts, we propose a novel logit under-normalization method in order to prioritize emissions of big blanks. With experiments on multiple languages and datasets, we show that multi-blank RNN-T methods could bring relative speedups of over +90%/+139% to model inference for English Librispeech and German Multilingual Librispeech datasets, respectively. The multi-blank RNN-T method also improves ASR accuracy consistently. We will release our implementation of the method in the NeMo ([this https URL](https://github.com/NVIDIA/NeMo)) toolkit.



 ## Authors



Hainan Xu (NVIDIA)

Fei Jia (NVIDIA)

Somshubra Majumdar (NVIDIA)

Shinji Watanabe (Carnegie Mellon University)

Boris Ginsburg (NVIDIA)

 

 

 ## Publication Date



Friday, November 4, 2022

 

 ## Research Area



[Speech Processing](/research-area/speech-processing)

 

 

 ## External Links



[Paper](https://arxiv.org/abs/2211.03541)