Zero resource code-switched speech challenge

This is the baseline system for the benchmark proposed in this ICASSP 2024 paper Zero Resource Code-Switched Speech Benchmark Using Speech Utterance Pairs for Multiple Spoken Languages.

Download code-switched zero resource data

Run this python code to download only the testing data.

# pip install datasets
from datasets import load_dataset
dataset = load_dataset("kph68/cs_zerospeech")

The full code-switch dataset is available at huggingface

Download monolingual data

# Spanish
wget https://dl.fbaipublicfiles.com/mls/mls_spanish.tar.gz

# French
wget https://dl.fbaipublicfiles.com/mls/mls_french.tar.gz

# Chinese
wget https://www.openslr.org/resources/68/train_set.tar.gz
wget https://www.openslr.org/resources/68/dev_set.tar.gz

# English
wget https://www.openslr.org/resources/12/train-clean-100.tar.gz
https://www.openslr.org/resources/12/dev-clean.tar.gz

# Extract
tar zxvf *.tar.gz

Data samples

zh-en	transcript	audio
correct	他们自称 artist.	http://sndup.net/kbjv
wrong	他们自称 tasty.	http://sndup.net/gxdk

es-en	transcript	audio
correct	En el presente, los interactive kiosks often have touch screens.	http://sndup.net/w2tj
wrong	En el presente, los interactive pantallas often have touch kiosks.	http://sndup.net/zjxm

fr-en	transcript	audio
correct	Ce dernier has evolved tout au long de l'histoire romaine.	http://sndup.net/kpms
wrong	Ce dernier has evolved tout during la long de l'histoire romaine.	http://sndup.net/x39f

Training

Train the K-means

Please modify the config files first.

python scripts/cpc/criterion/clustering/clustering_script.py \
--config scripts/cpc/criterion/clustering/cluster_config.yaml

Quantize audio & Deduplication & Preprocess (binarize, building dictionary)

Please modify the config files, and modify the paths in quantize_dedup.sh.

bash quantize_dedup.sh

Training the unit-LM

Please modify the parameters and some paths in train_unit.sh if needed.

bash train_unit.sh

Testing

Compute the pseudo-probability

Please modify the exp name and some paths in evaluate.sh first.

bash evaluate.sh

Citation

If you find this dataset or benchmark useful, please consider citing the following papers:

@INPROCEEDINGS{10446737,
  author={Huang, Kuan-Po and Yang, Chih-Kai and Fu, Yu-Kuan and Dunbar, Ewan and Lee, Hung-Yi},
  booktitle={ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={Zero Resource Code-Switched Speech Benchmark Using Speech Utterance Pairs for Multiple Spoken Languages}, 
  year={2024},
  volume={},
  number={},
  pages={10006-10010},
  keywords={Speech coding;Benchmark testing;Signal processing;Linguistics;Acoustics;Speech processing;Task analysis;Code-switch;Multilingual;Discrete unit;Zero resource;Self-supervised},
  doi={10.1109/ICASSP48485.2024.10446737}}

@article{yang2023investigating,
  title={Investigating Zero-Shot Generalizability on Mandarin-English Code-Switched ASR and Speech-to-text Translation of Recent Foundation Models with Self-Supervision and Weak Supervision},
  author={Yang, Chih-Kai and Huang, Kuan-Po and Lu, Ke-Han and Kuan, Chun-Yi and Hsiao, Chi-Yuan and Lee, Hung-yi},
  journal={arXiv preprint arXiv:2401.00273},
  year={2023}
}

nobel861017 / cs_zs_baseline