Librispeech: An ASR corpus based on public domain audio books

Author

Panayotov, Vassil ; Guoguo Chen ; Povey, Daniel ; Khudanpur, Sanjeev

Author_Institution

Center for Language & Speech Process., Johns Hopkins Univ., Baltimore, MD, USA

fYear

2015

fDate

19-24 April 2015

Firstpage

5206

Lastpage

5210

Abstract

This paper introduces a new corpus of read English speech, suitable for training and evaluating speech recognition systems. The LibriSpeech corpus is derived from audiobooks that are part of the LibriVox project, and contains 1000 hours of speech sampled at 16 kHz. We have made the corpus freely available for download, along with separately prepared language-model training data and pre-built language models. We show that acoustic models trained on LibriSpeech give lower error rate on the Wall Street Journal (WSJ) test sets than models trained on WSJ itself. We are also releasing Kaldi scripts that make it easy to build these systems.

Keywords

natural language processing; speech recognition; ASR corpus; Kaldi scripts; LibriSpeech corpus; LibriVox project; WSJ; Wall Street Journal; acoustic models; evaluating speech recognition systems; frequency 16 kHz; language-model training data; pre-built language models; public domain audio books; read english speech; training speech recognition systems; Bioinformatics; Blogs; Electronic publishing; Genomics; Information services; Resource description framework; Corpus; LibriVox; Speech Recognition;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on

Conference_Location

South Brisbane, QLD

Type

conf

DOI

10.1109/ICASSP.2015.7178964

Filename

7178964