Set of exercises on automatic speech processing.
Create and activate a virtual environment, for example with conda:
conda create -n spe python=3.10
conda activate spe
Clone this repository:
git clone https://github.com/idiap/idiap_spe.git
Install the idiap_spe
package in editable mode, so that you are free to make
any changes to the code:
cd idiap_spe
pip install -e .
The following exercises are available as both Jupyter and Org Babel notebooks.
- Speech signal analysis (Jupyter, Org)
- Hidden Markov models (Jupyter, Org)
- Grapheme-to-phoneme (G2P) conversion (Jupyter, Org)
A Jupyter notebook can be launched like this:
conda activate spe
jupyter notebook jupyter-notebooks/hmm.ipynb
Note: For the HMM exercise you might first need to install Graphviz following the instructions for your operating system here: https://graphviz.org/download/
The speech signal analysis and HMM exercises were originally developed in Matlab by Sacha Krstulović, Hervé Bourlard, and Mathew Magimai-Doss for the Speech Processing and Speech Recognition course at École polytechnique fédérale de Lausanne (EPFL).