Tag
This paper evaluates four leading real-time voice AI systems (GPT Realtime 2, Gemini 3.1 Flash Live, Qwen3.5 Omni Plus, Omni Flash) and finds they consistently act on words rather than vocal tone, ignoring distress, fear, or sarcasm even when they can perceive them—termed the 'emotional intelligence gap' of voice AI.
This paper presents an in silico simulation of the RAMPHO episodic buffer using phonetic entropy from wav2vec 2.0 to dissociate informational and energetic masking in multi-talker environments, revealing a cognitive-acoustic Pareto optimization problem.