A Roadmap for Privacy Preserving Speech Processing

This paper presents an overview of a strategy for enabling speech recognition to be performed in the cloud whilst preserving the privacy of users. The strategy advocates a demarcation of responsibilities between the client and server-side components for performing the speech recognition task. On the client-side resides the acoustic model, which symbolically encodes the audio and encrypts the data before uploading to the server. The server-side then employs searchable encryption-based language modelling to perform the speech recognition task. The paper details the proposed client-side acoustic model components, and the proposed server-side searchable encryption which will be the basis of the language modelling. Some preliminary results are presented, and potential problems and their solutions regarding the encrypted communication between client and server are discussed. Preliminary benchmarking results with acceleration of the client and server operations with GPGPU computing are also presented