Keeping Gadgets from Interrupting
Technology detects conversations while maintaining privacy
SOURCE: “Conversation Detection and Speaker Segmentation in Privacy-Sensitive Situated Speech Data”
Tanzeem Choudhury et al.
Interspeech 2007, August 27-31, Antwerp, Belgium
RESULTS: Researchers have developed software that can determine when a conversation is occurring and who is speaking. The voice data collected cannot be reconstructed into intelligible speech, so the system maintains a certain level of privacy.
WHY IT MATTERS: As gadgets such as cell phones become more prevalent, they are constantly interrupting people at inappropriate times. Engineers are interested in building context-aware devices that can determine when it’s suitable to notify users that someone is trying to contact them. One approach is to let a gadget with specialized software “listen” to conversations and decide whether it should interrupt.
METHODS: The researchers used a wearable microphone to collect conversation data from a group of 24 people over a span of 4,400 hours. As the audio data was collected, it was immediately processed so that only features such as the presence of speech and its rate, volume, pitch, and tone could be estimated. From such scant data, the verbal content of a conversation can’t be reconstructed. To determine who was speaking at any point in the conversation, the researchers used algorithms that looked for changes in these features and pauses in speech.
NEXT STEPS: The researchers plan to analyze more interactions to see how accurately they can determine whether, say, someone is speaking with a boss or a friend.