ClueCon Weekly with Andreas Granig: The 9 Dimensions of Voice Bot Performance

Host Luca sits down with Andreas Granig — founder and CEO of Sipfront, co-founder of SIPWISE, and one of the original contributors to the open-source VoIP ecosystem — for a deep technical dive into what it actually takes to build and operate a reliable voice bot in production.

Andreas walks through his framework of nine dimensions of voice bot performance: SIP performance, RTP performance, audio quality, audio performance, audio stability, audio correctness, content security, cost protection, and regulatory compliance. He covers why the traditional telephony foundation is now just the baseline, how to measure audio quality without a reference signal using referenceless models, the 200ms latency myth, why context pollution slows your bot down, and how fraudulent traffic can rack up tens of thousands of dollars in LLM costs over a weekend if you're not careful.

He also gets into jailbreaking via poetry (yes, really), the EU AI Act coming into force in August, GDPR obligations around voice transcription and recording disclosure, and why LLMs being nondeterministic means manual QA is no longer enough.