The first aibo took commands as musical tones. A year later, the ERS-210 recognized around 50 spoken words and learned its own name.
Sony researchers and Inria release an open-source model that splits a song into vocals, drums, bass and the rest.
Voice commands on PS5, first in beta, then a preview limited to US and UK accounts in English.
A leaked prototype runs Whisper, GPT-4, Llama 3 and Sony's own voice synthesis. Engadget called the voice robotic.
Sony AI shows how to control emotion intensity in a converted voice, aimed at dubbing and audiobooks (AAAI 2025).
The Tokyo lab behind Walrus publishes a speech-to-speech translation model. Walrus reuses its speech decoder.
One 8B model that transcribes, translates, speaks and answers questions about audio. Paper and demo only.