===============================================================================
README - NCHLT Speech Corpus pre-release 
===============================================================================

Version: clean v0.1 pre-release

IMPORTANT: Note that this is a pre-release of the National Centre for Human
Language Technologies (NCHLT) Speech Corpus.  The final version of the corpus
will be made available at http://http://rma.nwu.ac.za/. Any queries with regard
to updated versions can be directed to rma@nwu.ac.za.

This corpus is shared under the Creative Commons Attribution 3.0 license. 
For more information see license.txt in this directory.

When using this corpus, please cite:

- C. van Heerden, M.H. Davel and E Barnard, "The semi-automated creation of
  stratified speech corpora", Pattern Recognition Association of South Africa
  (PRASA), Johannesburg, December 2013.

===============================================================================

Additional documentation:

- N.J. de Vries, M.H. Davel, J. Badenhorst, W.D. Basson, F. de Wet, E. Barnard
  and A. de Waal, "A smartphone-based ASR data collection tool for under-
  resourced languages", Speech Communication, Volume 56, January 2014, 
  pp 119–131.

- C van Heerden, M.H. Davel and Barnard, E., "Medium-Vocabulary Speech
  Recognition for Under-Resourced Languages", 3rd International Workshop on
  Spoken Language Technologies for Under-resourced Languages (SLTU 2012), Cape
  Town, South Africa, May 2012, pp. 146-151.

===============================================================================
LANGUAGE INFORMATION
===============================================================================

A separate speech corpus exists for each of South Africa's eleven official 
languages. Each of the languages is represented by its three-character 
ISO 639-3:2007 language code, as listed below: 

---------------------------------
| Language code | Language name |
---------------------------------
| afr           | Afrikaans     |
| eng           | English       |
| nbl           | isiNdebele    |
| xho           | isiXhosa      |
| zul           | isiZulu       |
| nso           | Sepedi        |
| tsn           | Setswana      |
| sot           | Sesotho       |
| ssw           | Siswati       |
| ven           | Tshivenda     |
| tso           | Xitsonga      |
---------------------------------

===============================================================================
DIRECTORY AND DATA FORMAT DESCRIPTION
===============================================================================

The corpus for each language consists of audio and transcriptions in the 
following directory structure: 

 nchlt_<language_code>
  |_ audio
    |_ <spk_id>
      |_ *.wav
  |_ transcriptions
    |_ nchlt_<language_code>.trn.xml
    |_ nchlt_<language_code>.tst.xml

where <language_code> and <spk_id> are used as placeholders for the language 
and speaker identifiers, respectively. 

Transcriptions are provided in XML (UTF8) format and subdivided into individual
train and test suites ("nchlt_<language_code>.trn.xml" and 
"nchlt_<language_code>.tst.xml", respectively). 

Each XML entry contains the orthographic transcription of the specific 
utterance as well as additional metadata: the age and gender of the speaker, 
the utterance duration and the md5sum of the audio file. 

Individual recordings are encoded using mono 16-bit signed integer PCM sampled 
at 16kHz. Recordings are subdivided using a unique speaker identifier 
(<spk_id>) for every speaker. Speaker identifiers for the test suite audio 
correspond to the integer values of 500-507.

===============================================================================
CORPUS STATISTICS
===============================================================================

Language     speakers	female/male	duration

Afrikaans	210	107/103		56:22
English		210	100/110		56.26
isiNdebele	148	78/70		56.14
isiXhosa	209	106/103		56.15
isiZulu		210	98/112		56.15
Sepedi		210	100/110		56.20
Sesotho		210	113/97		56.19
Setswana	210	109/101		56.20
Siswati		197	96/101		56.15	
Tshivenda	208	83/125		56.17
Xitsonga	198	95/103		56.16 	

===============================================================================