Apple uses federated learning to refine its voice assistant without transferring raw data from users' phones, showing promise for privacy-related use cases across industries

December 18, 2019 | Case Study

Modern voice assistants rely on massive amounts of data to train the machine learning models that power their speech recognition and natural language understanding features. Given the heated race between tech giants like Amazon, Google, and Apple to develop the most capable voice assistants, there's a strong appetite to collect more user data to do so. However, following several incidents of employees listening to voice recordings from smart devices in addition to countless high-profile data breaches, consumers and regulatory bodies are demanding greater data privacy. This has put tech companies in a conundrum where they must decide between maintaining user privacy and falling behind in the voice assistant race, as Apple has already done.