Ameca

This Ameca demo uses GPT-3 for conversation and translation. DeepL is used for language detection. The voices are Amazon Polly Neural voices. They are working on a demo using ElevenLabs voice cloning, it adds some complexity because of the additional phoneme and viseme generation for lip sync. All of this is put together on their Tritium software platform, they will be releasing a beta public version of in the coming months. It includes virtual Ameca and support for importing other robot models in SDF format.

The makers noticed that the processing time with GPT-4 was much longer than GPT-3 and made Ameca appear less responsive with her facial expressions.