Hi folks, Alex here from ThunderPhone. Today we're launching v2. Voice AI has been a hot topic for the past year or two, both for automating phone calls and for adding voice features to apps. Most of the voice agents deployed have been using a "3 step pipeline": a transcription model to turn user speech into text, an LLM to generate a response in text, and a TTS model to turn the response text into audio. But the performance of these 3-step voice agents has often been so-so, largely for 3 reason...