The avatars normally use our Text To Speech support for speech (or native TTS in HTML5, Android, and iOS).
You can also associate audio files for actions and poses.
We do not have support for associating real speech audio files with responses yet, but that is something we plan on. Currently if you have your own speech audio files you can use our SDK to have an avatar simulate speaking them. You can copy the SDK source code that uses our TTS audio and insert your own speech audio. You will need to code this yourself.