QSpeech: Low-Qubit Quantum Speech Application Toolkit

Accepted by the International Joint Conference on Neural Networks (IJCNN) 2022

This repository is the official implementation of QSpeech: Low-Qubit Quantum Speech Application Toolkit.

It proposes:

• The low-qubit variational quantum circuit (VQC).
• A library for the rapid prototyping of hybrid quantum-classical neural networks in speech applications.

Low-qubit VQC	QSpeech Framework

For the hybrid quantum-classical neural networks in speech applications, we implement the Quantum M5(QM5), Quantum Tacotron(QTacotron) and Quantum Transformer-TTS(QTransformer-TTS).

• Linux (Test on Ubuntu18.04)
• Python3.6+ (Test on Python3.6.8)
• PyTorch
• PennyLane
• Librosa (version 0.7.2)
• Numba (version 0.48.0)

• QCircuit: the variational quantum circuit(VQC) and low-qubit VQC.
• QLayer: the qlstm, qgru, qattention, qconv.
• QModels: the qm5, qtransformer, qtacotron.

The code of qlstm and qtransformer are based on these two projects as follow:

• qlstm
• qtransformer

• LJSpeech1.1
• SpeechCommandV0.02

• cd ./Examples/QM5
• Modify the config.py, like the path of dataset
• python3 speech-command-recognition.py

• cd ./Examples/QTacotron
• Modify the hyperparams.py, like the path of dataset
• python3 train.py --batch_size 2

• cd ./Examples/QTransformerTTS
• Modify the config.py, like the path of dataset
• python3 train.py

If you find QSpeech useful in your research, please consider citing:

@inproceedings{hong2022qspeech,
  title={QSpeech: Low-Qubit Quantum Speech Application Toolkit},
  author={Hong, Zhenhou and Wang, Jianzong and Qu, Xiaoyang and Zhao, Chendong and Tao, Wei and Xiao, Jing},
  booktitle={2022 International Joint Conference on Neural Networks (IJCNN)},
  pages={1--8},
  year={2022},
  organization={IEEE}
}