Sesame (https://www.sesame.com/) is an oral conversation model currently in research version, based on the open-source csm model (https://github.com/SesameAILabs/csm).
Current research is documented: https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice. Two voices are available for testing: https://app.sesame.com/ before the launch of a dedicated application (waiting list for access to the beta version).

Based on Gemma 3 and CSM, 2 voices are available, Maya and Miles, in 140 languages. The model under test is connected to the web for real-time searches.

Use with and without an account: 5 minutes of conversation possible without an account and 30 minutes with an account.

Note: with accounts the conversations are used by default to train the model, a case is to be unchecked in the preferences.

Free and partly open-source.