The package contains Fongbe speech corpus with audio data in the directory data/. The data directory contains 3 subdirectories:
- train - speech data and transcription for training automatic speech recognition system (Kaldi ASR format [1])
- test - speech data and transcription (verified) for testing the ASR system (Kaldi ASR format)
- local - for now, contains the Fongbe vocabulary. Once you will ran the run.sh script it will contain the dict/ and lang/ directories needed to build the ASR system.
LM/ directory contains a text corpus and the language model.
More details on the corpus and how it was collected can be found on the following publication (please cite this bibtex if you use this data).
@inproceedings{laleye2016FongbeASR,
title={First Automatic Fongbe Continuous Speech Recognition System: Development of Acoustic Models and Language Models},
author={A. A Laleye, Fréjus and Besacier, Laurent and Ezin, Eugène C. and Motamed, Cina},
year={2016},
organization={Federated Conference on Computer Science and Information Systems}}
In kaldi-scripts/ you will find:
- 00_init_paths.sh - it initializes your PATH variable (NOTE: you have to modify this file by yourself)
- 01_init_symlink.sh - it creates the symbolic links required to run the Kaldi scripts
- 02_lexicon.sh - it creates the dict/ directory used by Kaldi
- 03_lm_preparation.sh - it creates the lang/ directory used by Kaldi
| Acoustic models | WER score (%) on test |
|---|---|
| Monophone (13 MFCC) | 35.34 |
| Triphone (13 MFCC) | 27.42 |
| Triphone (13 MFCC + delta + delta2) | 26.75 |
| Triphone (39 features) + LDA and MLLT | 22.25 |
| Triphone (39 features) + LDA and MLLT + SAT and FMLLR | 17.77 |
| Triphone (39 features) + LDA and MLLT + SGMM | 16.57 |
[1] KALDI: http://kaldi.sourceforge.net/tutorial_running.html