Deepgram’s TTS product and research teams show you how to define scoring criteria, automate what you can measure, and structure human eval that scales.

Date: Monday, May 5, 2026

Time: 10:00 AM PT | 1:00 PM ET

Duration: 60 minutes

Format: Live Technical Webinar

Jeff Liu, Product Manager, TTS, Deepgram

Constantin Weisser, Research Scientist, Deepgram

Moderator: Hasan Jilani

Evaluating TTS in a voice agent is a different problem than evaluating STT. There’s no single metric like word error rate. You’re dealing with multiple quality dimensions, and none of them have standard definitions or measurement methods.

Most teams default to listening and deciding if it sounds “good enough.” That works early on. It breaks down when you need to make real decisions across models, voices, and use cases.

In this session, Deepgram’s TTS product and research teams walk through how to build an evaluation pipeline you can actually use: how to define scoring criteria tied to your use case, which metrics you can automate, and how to structure human evaluation with multiple raters and rubrics.

• A scoring rubric tied to your use case: How to define what “good TTS” means for your product before you start testing, with specific evaluation dimensions and criteria.
• An automated metrics pipeline: Which TTS qualities you can measure programmatically, which tools to use, and where automated metrics fall short.
• A structured human eval workflow: How to move from one person listening to multi-rater evaluation with rubrics, scoring scales, and inter-rater reliability.
• A path from model eval to production: How to connect model-level scores to the question that matters: will real users trust this voice?

Engineers and product teams shipping voice agents. If you’re comparing TTS models, defining voice requirements, or need an evaluation process that works at scale, this is for you.

Register to get a complete TTS evaluation workflow you can take back to your team and use immediately.