The source code and audio datasets of my PhD project.1. https://www.openslr.org/12LibriSpeech is a corpus of approximately 1000 hours of 16kHz read English speech, prepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read audiobooks from the LibriVox project, and has been carefully segmented and aligned.Acoustic models, trained on this data set, are available at kaldi-asr.org and language models, suitable for evaluation can be found at http://www.openslr.org/11/.For more information, see the paper "LibriSpeech: an ASR corpus based on public domain audio books", Vassil Panayotov, Guoguo Chen, Daniel Povey and Sanjeev Khudanpur, ICASSP 20152.https://www.openslr.org/17MUSAN is a corpus of music, speech, and noise recordings.This work was supported by the National Science Foundation Graduate Research Fellowship under Grant No. 1232825 and by Spoken Communications.You can cite the data using the following BibTeX entry:@misc{musan2015, author = {David Snyder and Guoguo Chen and Daniel Povey}, title = {{MUSAN}: {A} {M}usic, {S}peech, and {N}oise {C}orpus}, year = {2015}, eprint = {1510.08484}, note = {arXiv:1510.08484v1}}3. source_code.zipThe program from parts of my PhD project.4.SJ_EXP.zipThe program of the subjective experiment corresponding to the last chapter.

Introduction

Fisher and CALLHOME Spanish-English Speech Translation was developed at Johns Hopkins University and contains English reference translations and speech recognizer output (in various forms) that complement the LDC Fisher Spanish (LDC2010T04) and CALLHOME Spanish audio and transcript releases (LDC96T17). Together, they make a four-way parallel text dataset representing approximately 38 hours of speech, with defined training, development, and held-out test sets.

Data

The source data are the Fisher Spanish and CALLOME Spanish corpora developed by LDC, comprising transcribed telephone conversations between (mostly native) Spanish speakers in a variety of dialects. The Fisher Spanish data set consists of 819 transcribed conversations on an assortment of provided topics primarily between strangers, resulting in approximately 160 hours of speech aligned at the utterance level, with 1.5 million tokens. The CALLHOME Spanish corpus comprises 120 transcripts of spontaneous conversations primarily between friends and family members, resulting in approximately 20 hours of speech aligned at the utterance level, with just over 200,000 words (tokens) of transcribed text.

Translations were obtained by crowdsourcing using Amazon's Mechanical Turk, after which the data was split into training, development, and test sets. The CALLHOME data set defines its own data splits, organized into train, devtest, and evltest, which were retained here. For the Fisher material, four data splits were produced: a large training section and three test sets. These test sets correspond to portions of the data where four translations exist.

Samples

Please view this corpus and mapping sample.

Updates

None at this time.

Portions © 2014 Chris Callison-Burch, Matt Post, © 1996, 1997, 2005-2006, 2010, 2014 Trustees of the University of Pennsylvania