Voice modal has worked for us, we were using Cartesia, changing to openai worked.