MixedVoices: Voice Agent Analytics

Learn how to use a Python API and Streamlit dashboard to track, visualize, and test voice agents through flowchart simulations, version control, and edge‑case analysis.

Overview

MixedVoices is an open source analytics and testing platform for voice agents. It helps builders track, visualize, and optimize performance by view conversation flowcharts, running simulations and suggesting prompt improvements.

Technical Details:

LLM chain of thought to break calls into reusable and deterministic steps
Diarization of audio: Still not solved, accuracy of current solutions is still subpar, using stereo audio instead
Emotion analysis: gpt4o (hume planned later)
Using flowcharts to condense 1000s of recordings and create simulations that cover all edge cases
Creating simulations to test agent on, ensuring entire graph is covered

Demo Structure:

My experience in building Voice agents and facing problems due to unreliability
Project walk through: Python API (Few lines of code to integrate) + Streamlit Dashboard (Runs locally)
Importance of Version control to track agent evolution over time
Examples of edge cases discovered
Examples of simulations and how they help in continuous evaluation

Links

Tech stack