According to MIT Technology Review, a paper published on Friday (22nd) local time showed thatappleThe company's researchers are exploringThe possibility of using artificial intelligence to detect when a user is talking to a device such as an iPhone, thereby eliminatingSiri” The technical requirements for trigger phrases like this.
In the study, which was uploaded to Arxiv and has not yet been peer-reviewed, researchersA large language model was trained using speech captured by a smartphone as well as acoustic data from background noise., to look for patterns that “may indicate that the user needs assistance from the device.”
The paper states that the model is partly based on OpenAI's GPT-2.Because it is relatively lightweight, it can run on devices such as smartphonesThe paper also describes more than 129 hours of data and additional text data used to train the model, but does not specify the source of the recordings used in the training set. According to LinkedIn profiles, six of the seven authors list their affiliation as Apple, three of whom work on Apple's Siri team.
The paper’s conclusions are “encouraging,” claiming that the model is able to make more accurate predictions than audio-only or text-only models, and that it improves further as the model scales.
Currently, Siri's functionality is achieved by retaining a small amount of audio.It won't start recording or prepare to answer user prompts until it hears a trigger phrase like "Hey, Siri".
Jen Kim, a privacy and data policy researcher at the Stanford Institute for Human-Centered AI, said removing the “Hey, Siri” prompt could increase concerns about devices “always listening.”