An intelligent virtual assistant (IVA) is an artificial intelligence system that emulates a human person in terms of intelligence, understanding, responsiveness and adaptability. An IVA differs from traditional interactive voice response systems largely in its use of machine learning, which allows an IVA to predict user actions. The global VIA market is expected to reach approximately US$45 billion by 2027, expanding at a compound annual growth rate (CAGR) of 34%. Much of the IVA market is owned by Amazon and Google, although Apple has indicated plans to launch a desktop IVA product to compete with Amazon’s Echo.
Virtual assistant trends indicate a growing desire among users to interact first through voice, that’s to say, interact with a device primarily by voice. Voiceprint recognition and audio processing before waking up are therefore at the heart of the next generation of IVA. Big tech patent filings in recent years demonstrate substantial advances in IVA technology, while raising significant privacy concerns.
In 2018, Amazon got a patent directed towards “the vocal determination of the physical and emotional characteristics of users”. According to the patent, the IVA will be able to “determine one or more voice characteristics based at least in part on a user’s speech or voice input” such as “a user’s gender, age, or gender range.” age of the user, ethnicity or language accent of the user, emotion of the user, background noise from the environment in which the user is located, and other voice characteristics. the IVA identifies a physical or emotional characteristic, it generates a label for that characteristic and links the label to a data file of the voice input.Tags based on physical or emotional characteristics will allow the IVA to suggest a targeted content to a user. For example, if the IVA detects that a user seems bored, the IVA can ask “are you in the mood for a movie?” Ultimately, this would allow companies to deliver targeted advertisements based on the an individual’s mood at a given time.
In 2019, Amazon also received a patent directed to “pre-wake speech processing”. A “wake word” is the wake command that activates the IVA, for example, “Alexa” or “Hey Alexa”. Enabling the IVA’s pre-wake word would in theory allow a user to make a request using more natural language, such as “play Alexa music” versus “Alexa, play music.” However, this feature requires further audio processing: according to Amazon’s 2019 patent, “[a]When the device detects audio, it can process the audio (before or after the audio is stored in the buffer) to determine if the audio includes a wake word. The device can continue to do this until it detects a wake word in the received audio. In other words, Echo would store and process all the audio it hears and perform an ex post facto assessment of whether that audio included a command.
Apple has taken significant steps to develop its own IVA products and systems. For example, in February 2022, Apple filed a first patent application for a “voice-activated electronic device” and a second application directed to a “desktop electronic device”, apparently to better compete with the Amazon Echo and Google Assistant. Building on its previous text-to-speech systems (for example, this patent for methods to infer user intent from speech), Apple recently filed a patent application aimed at improving “methods and user interfaces for voice control of electronic devices”.
While these developments in IVA voice recognition and analysis improve the user experience, they also introduce a number of potential privacy issues. Pre-wake speech processing can allow the IVA to analyze and store significantly more audio than the user intended. It’s unclear exactly how much data is collected and stored through smart home devices, and how long the tech giants retain that data. Additionally, while the collection and use of voiceprint data was anticipated, the implementation of targeted advertising based specifically on emotional or physical conditions represents an unprecedented degree of commercialization of personal information. As the technology supporting digital assistants continues to improve, legislators need to adapt existing privacy laws accordingly, and consumers need to be aware of the extent to which their data is being used.