MOCHA-TIMIT

General

Authors: Alan Wrench, Queen Margaret University College

Funded by: Engineering and Physical Sciences Research Council:

When created: November 1999

Availability: English speakers available here free for non-commercial use and may be distributed on CDROM for a fee.

Purpose: Phonetically balanced dataset for training an automatic speech recognition system

Description: Overview

Instrumentation

Components:

• Microphone 16kHz sample rate (audio-technica ATM10a)
• Laryngograph 16kHz sample rate
• Electromagnetic Articulograph 500Hz sample rate (Carstens 10 Channel)

  upper incisor

  lower incisor

  upper lip

  lower lip

  tongue tip

  tongue blade

  tongue dorsum

  velum

  • EPG 200Hz sample rate
  • SVHS video of front view of mouth area. (Available by special request)

  Corpus

  Texts:

  1. Sentences:
     • A set of 460 sentences designed to include the main connected speech processes in English (eg. assimilations, weak forms ..).
     Orthography

  Subjects: 2 speakers, 1 male and 1 female are currently available but another 38 are planned to be completed by May 2001. The subjects have a variety of accents of English.

  Conditions:

  All recordings made in the same sound damped studio at the Edinburgh Speech Production Facility. All data were recorded direct to computer and carefully synchronised.

  Languages: English

  Access

  Platforms: The data files have headers which retain byte order information.

  Media: Internet (FTP) and possibly CDROM

  SVHS video available by special request.

  Format:

  Audio and Laryngograph are stored with 1024 byte ascii NIST headers. EPG remains in raw binary (8 bytes per sample). EMA data is stored in Edinburgh Speech Tools Trackfile format consisting of a variable length ascii header and a 4 byte float representation per channel. The first channel is a time value in seconds the second value is always 1 (used to indicate if the sample is present or not) subsequent 5 values are coil 1-5 x-values followed by coil 1-5 y-values followed by coil 6-10 x-values and finally coils 6-10 y-values.

  Size:~200kBytes per speaker

  Software:

  • Edinburgh Speech tools is free and contains routines ch_wave and ch_track which can be used to convert the waveform and EMA files into other formats such as ESPS waves format, HTK or raw binary as well as other routines such as pitch tracking.
  • MATLAB - a set of macros to read and write the stored data formats are available along with supplimentary routines for the EMATOOLS set.

  Software Documentation:

  Edinburgh Speech Tools

  MATLAB-EMATools

  Software Source/Executables:

  See above.

  Further information

  Contact: A. Wrench

  Address:

  Dept. Speech and language Sciences, Queen Margaret University College, Clerwood Terrace, Edinburgh. EH12 8TS.

  Telephone:

  +44 131 317 3692

  Fax:

  +44 131 317 3689

  Email:

  a.wrench@sls.qmced.ac.uk

  WWW:

  http://sls.qmuc.ac.uk