# diSpeech

**Repository Path**: mirrors_Orange-OpenSource/diSpeech

## Basic Information

- **Project Name**: diSpeech
- **Description**: Materials to generate diSpeech datasets, composed of phonemes generated with Klatt synthetizer, for speech disentanglement purposes.
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2021-10-22
- **Last Updated**: 2026-01-03

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# diSpeech
Materials to generate diSpeech datasets for speech disentanglement purposes. Klatt synthetizer is used to generate phonemes.

## Files
- `dataset.py`: main classes to use to generate and load diSpeech datasets.
- `utils.py`: usefull functions to synthetize phoneme waveforms and compute log mel-filter bank from waveforms.
- `gen_*.py`: scripts to generate mentioned datasets in *diSpeech : A synthetic toy dataset for speech disentangling* (Table 2).
- `tdklatt.py`: Klatt synthetizer implementation, from [https://github.com/guestdaniel/tdklatt](https://github.com/guestdaniel/tdklatt).
- `vowel_formants.csv`: reference formants of french vowels.