For running this experiment we used a Linux environment with Python 3.6.
You can see a list of python dependencies at requirements.txt.
To install it on conda virtual environment (conda install --file requirements.txt).
To install it on pip virtual environment (pip install -r requirements.txt).
To run it on TIMIT dataset we have first to pre-process the data, removing the start and ending silences moments and also normalizing the audio sentences.
python TIMIT_preparation.py $TIMIT_FOLDER $OUTPUT_FOLDER data_lists/TIMIT_all.scp
where:
- $TIMIT_FOLDER is the folder of the original TIMIT corpus
- $OUTPUT_FOLDER is the folder in which the normalized TIMIT will be stored
- data_lists/TIMIT_all.scp is the list of the TIMIT files used for training/test the speaker id system.
then, we can run the experiment itself by typing.
python speaker_id.py --cfg=cfg/$CFG_FILE
where:
- $CFG_FILE is the name of the cfg configuration file which is located at cfg folder.
We have made available several cfg configuration files for the experiments, if you want to run the experiment with the AM-MobileNet1D you must use the AM_MobileNet1D_TIMIT.cfg file, otherwise you can use the AM_MobileNet_XXX.cfg file where the XX refers to the dataset name.
When training have a look at the cfg configuration file, the output paths for the model and the result (res.res) files are placed there.
We have also made available some results from our experiments, you can check them at exp folder. The resume of the results are saved in the res.res files.
If you use this code or part of it, please cite us!
@misc{nunes2020ammobilenet1d,
title={AM-MobileNet1D: A Portable Model for Speaker Recognition},
author={João Antônio Chagas Nunes and David Macêdo and Cleber Zanchettin},
year={2020},
eprint={2004.00132},
archivePrefix={arXiv},
primaryClass={cs.SD}
}
You can also find the preprint at arXiv.