Voice agents combine speech recognition, a language model and speech synthesis into one conversation. Each part can look fine on its own while the whole experience falls apart: a misheard account number, a mispronounced street name, an awkward pause or an agent that talks over the caller.

Evaluation is at the heart of my work. Through my research and my role as Technical Program Chair of Interspeech 2025, I keep close track of how the field measures speech quality, and I translate that into practical tests for your product.

What I offer

  • End-to-end quality audits: structured testing of your agent across realistic scenarios, with a prioritized report.
  • Hard-case test suites: names, addresses, numbers, spelling, accents and code-switching.
  • Conversational quality: latency, turn-taking, interruptions and prosody that fits the dialogue.
  • Evaluation design: combining listening tests, automatic metrics and production monitoring into a framework your team can keep using.