Privacy-Preserving Isolated-Word Recognition

Abstract

In this chapter we present a privacy-preserving framework for isolated word recognition based on Hidden Markov Models (HMMs). Classification based on HMMs is common in machine learning and is nearly ubiquitous in applications related to speech processing. We consider a multi-party scenario in which the data and the HMMs belong to different individuals and cannot be shared. For example, Alice wants to analyze speech data from telephone calls. She outsources the speech recognition task to Bob, who possesses accurate HMMs obtained via extensive training. Alice cannot share the speech data with Bob owing to privacy concerns while Bob cannot disclose the HMM parameters because this might leak valuable information about his own training database.