Improving LSTM-CTC based ASR performance in domains with limited training data

2017/07/03 by Jayadev Billa, Billa, Jayadev · 1 citation
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Music and Audio Processing #Natural Language Processing Techniques #Speech Recognition and Synthesis

paper · pdf · doi:10.48550/arxiv.1707.00722

openalex publication_date 2017/07/03 · openalex created_date 2017/07/14 · openalex updated_date 2026/07/28

Abstract

This paper addresses the observed performance gap between automatic speech recognition (ASR) systems based on Long Short Term Memory (LSTM) neural networks trained with the connectionist temporal classification (CTC) loss function and systems based on hybrid Deep Neural Networks (DNNs) trained with the cross entropy (CE) loss function on domains with limited data. We step through a number of experiments that show incremental improvements on a baseline EESEN toolkit based LSTM-CTC ASR system trained on the Librispeech 100hr (train-clean-100) corpus. Our results show that with effective combination of data augmentation and regularization, a LSTM-CTC based system can exceed the performance of a strong Kaldi based baseline trained on the same data.

Citations

• Kaldi Speech Recognition Toolkit
• EESEN: End-to-End Speech Recognition using Deep RNN Models and WFST-based Decoding
• Bidirectional recurrent neural networks
• Deep Speech 2: End-to-End Speech Recognition in English and Mandarin
• Recurrent Dropout Without Memory Loss

Cited by

• Random Bias Initialization Improves Quantized Training

Related

• End-To-End Speech Recognition Using A High Rank LSTM-CTC Based Model
• Relaxing the Conditional Independence Assumption of CTC-based ASR by Conditioning on Intermediate Predictions
• LV-CTC: Non-autoregressive ASR with CTC and latent variable models
• FastInject: Injecting Unpaired Text Data into CTC-based ASR training
• Intermediate Loss Regularization for CTC-based Speech Recognition