Federated Automatic Speech Recognition in Federated Ecology
He Hongling, Peng Ye, Fei-Yue Wang · 2021 IEEE 1st International Conference on Digital Twins and Parallel Intelligence (DTPI) · 2021
Automatic Speech Recognition (ASR) requires a lot of data for training, and more data is better than more advanced algorithms. However, the training data of speech recognition may expose personal information such as voice prints. We have heard some cases of fraud using synthetic voices. Existing federated learning methods have many limitations that restrict the further applications in large practical scenarios. In this paper, we propose a set of federated data methods to transfer training features, which eventually not only keep the convenience and effectiveness of ASR training by nature but also protect the speaker's personal information.