A comparison of end-to-end models for long-form speech recognition

• DOI: 10.48550/arxiv.1911.02242
• Publication Date: 2019-01-01
• Authors: Chung‐Cheng Chiu; Wei Han; Zhang Yu; Ruoming Pang; Sergey Kishchenko; Patrick Nguyen; Arun Narayanan; Hank Liao; Shuyuan Zhang; Anjuli Kannan; Rohit Prabhavalkar; Zhifeng Chen; Tara N. Sainath; Yonghui Wu
• Keywords: End-to-end principle

End-to-end automatic speech recognition (ASR) models, including both attention-based models and the recurrent neural network transducer (RNN-T), have shown superior performance compared to conventional systems. However, previous studies focused primarily on short utterances that typically last for just a few seconds or, at most, tens of seconds. Whether such architectures are practical long from minutes hours remains an open question. In this paper, we investigate improve end-to-end long-form transcription. We first present empirical comparison different real world task demonstrate RNN-T model is much more robust than systems in regime. next explore two improvements significantly its performance: restricting attention be monotonic, applying novel decoding algorithm breaks into shorter overlapping segments. Combining these improvements, show can very competitive recognition.