TTSLab Lwazi2 Afrikaans build scripts

This repository contains pre-setup data and scripts for building an Afrikaans TTS voice using TTSLab and associated build tools. This repository should be built against TTSLab commit 5cb979c.... The recommended way to do this is described in ./HOWTO.md and involves cloning this repository and running the accompanying Docker script (./Dockerfile).

The copyright and licence information for scripts in this repository can be found in ./COPYRIGHT and ./LICENCE-APACHE/./LICENCE-MIT. This repository also contains software and data developed by third parties used under licence in the ./src and ./recs directories. Copyright and licence information for third-party components is contained in each individual sub-directory or source file. Summaries of these components including attribution information are briefly presented below.

Since the speech corpus bundled here is specifically released under a licence that prohibits commercial use, this repository can be used as-is for research purposes or to demonstrate how to use the TTSLab toolkit to build voices (in academic work please cite 1).

Speech corpus and language data

Lwazi II Afrikaans TTS Corpus

The Lwazi 2 Afrikaans text-to-speech corpus was originally developed by the CSIR in South Africa and can be used and distributed under the terms of a Creative Commons Licence which prohibits commercial use. If this corpus is used for research purposes, please refer to the CSIR technical report (cite 1).

For more information and links to the original corpus distribution refer to ./recs/README.md.

Afrikaans Pronunciation Dictionary

The pronunciation dictionary used here was enhanced for use in TTS during the NTTS project funded by the Department of Arts and Culture (DAC) of the Government of South Africa and is available on Github under a Creative Commons licence (cite 2 or 3). The original automatic speech recognition (ASR) dictionary from which the current version was derived is available on SourceForge (cite 4).

Software

External

The source for the following external pieces of software are fetched automatically and compiled here (please refer to the original Github repositories for copyright and licence conditions):

1. The TTSLab toolkit and associated development tools (please cite 1).
2. The ZA_lex lexical resources for South African languages.
3. Praat (see also http://www.fon.hum.uva.nl/praat/).

Bundled

The following files contain source code from third parties with their own copyright and licence information:

1. Edinburgh Speech Tools: ./src/speech_tools-2.4-release.tar.gz
2. Speech Signal Processing Toolkit (SPTK): ./src/SPTK-3.8.tar.gz
3. HMM-based Speech Synthesis System (HTS): ./src/HTS-2.3beta_for_HTK-3.4.1.tar.bz2
4. Sequitur G2P: A trainable Grapheme-to-Phoneme converter (cite 5): ./src/g2p-r1668-r3.tar.gz

Not included

The source code for the following third-party tools are required but are not included here (this should be downloaded and placed in the correct locations before building as described in ./HOWTO.md:

1. The Hidden Markov Model Toolkit (HTK) version 3.4.1: ./src/HTK-3.4.1.tar.gz
2. HTK HDecode extension version 3.4.1: ./src/HDecode-3.4.1.tar.gz

References

1. K. Calteaux, F. de Wet, C. Moors, D. R. van Niekerk, B. McAlister, A. Sharma Grover, T. Reid, M. Davel, E. Barnard, and C. van Heerden, "Lwazi II Final Report: Increasing the impact of speech technologies in South Africa," Council for Scientific and Industrial Research, Pretoria, South Africa, Tech. Rep. 12045, Feb. 2013.

@techreport{calteaux2013lwazi2,
	author = {Calteaux, K. and de Wet, F. and Moors, C. and Van Niekerk, D. R. and McAlister, B. and Sharma Grover, A. and Reid, T. and Davel, M. and Barnard, E. and Van Heerden, C.},
	title = {{Lwazi II Final Report: Increasing the impact of speech technologies in South Africa}},
	number = {12045},
	institution = {{Council for Scientific and Industrial Research}},
	address = {Pretoria, South Africa},
	month = feb,
	year = {2013}
}

2. D.R. van Niekerk, "Syllabification for Afrikaans speech synthesis," in Proceedings of the 27th Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp 31-36, Stellenbosch, South Africa, December 2016.

@inproceedings{vniekerk2016prasa,
	title = {{Syllabification for Afrikaans speech synthesis}},
	booktitle = {{Proceedings of the Twenty-Seventh Annual Symposium of the Pattern Recognition Association of South Africa (PRASA)}},
	author = {van Niekerk, D. R.},
	address = {Stellenbosch, South Africa},
	month = dec,
	year = {2017},
	pages = {31-36}
}

3. D.R. van Niekerk, "Evaluating acoustic modelling of lexical stress for Afrikaans speech synthesis," in Proceedings of the 28th Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp 86-91, Bloemfontein, South Africa, December 2017.

@inproceedings{vniekerk2017prasa,
	title = {{Syllabification for Afrikaans speech synthesis}},
	booktitle = {{Proceedings of the Twenty-Seventh Annual Symposium of the Pattern Recognition Association of South Africa (PRASA)}},
	author = {van Niekerk, D. R.},
	address = {Stellenbosch, South Africa},
	month = dec,
	year = {2017},
	pages = {31-36}
}

4. M.H. Davel and F. de Wet, "Verifying pronunciation dictionaries using conflict analysis," In Proc. Interspeech, Makuhari, Japan, Sept 2010, pp 1898-1901.

@inproceedings{davel10verifying,
	authors = {M.H Davel and F. de Wet},
	title = {Verifying pronunciation dictionaries using conflict analysis},
	booktitle = {Proc. Interspeech},
	pages = {1898-1901},
	address = {Makuhari, Japan},
	year = {2010},
	month = {Sept.}
}

5. M. Bisani and H. Ney, "Joint-Sequence Models for Grapheme-to-Phoneme Conversion". Speech Communication, Volume 50, Issue 5, May 2008, Pages 434-451.

@article{bisani2008jsmg2p,
	title = {{Joint-Sequence Models for Grapheme-to-Phoneme Conversion}},
	volume = {50},
	issn = {0167-6393},
	number = {5},
	journal = {{Speech Communication}},
	author = {Bisani, Maximilian and Ney, Hermann},
	month = may,
	year = {2008},
	pages = {434--451}
}