Get Started
Home
Topics
Search
Library
Research questionHow can speech language models protect one user’s private context when responding to another speaker?In shared environments, a speech language model must determine whose information a request concerns before answering. Responses that ignore speaker identity can expose private details such as appointments to other users.
AI
Alignment & Safety
Audio & Speech
Audio & Speech Processing
Evaluation & Benchmarks
Latest papersRecent research connected to this question, newest first.VoxPrivacy: A Benchmark for Evaluating Interactional Privacy of Speech Language ModelsThe source evaluates nine speech language models using a 32-hour bilingual dataset and a human-recorded subset. It covers direct secrecy commands, conditional privacy decisions, and proactive privacy inference, and reports additional results after fine-tuning on a 4,000-hour training set.research paper · Sep 3, 2026
Related questions
How can vision-language models decide when missing user context requires deferring rather than answering?How can language models answer sensitive prompts helpfully without compromising safety?Can language agents maintain hidden state consistently across dialogue branches using only public conversation history?How can mixture-of-experts language models preserve safety when adversaries manipulate sparse expert routing?