Voice AI Pronunciation Challenges

​Join us for a deep dive into the challenges with pronunciation when deploying voice AI applications in production:

• ​Why TTS models struggle to pronounce certain words like Chipotle or Acetaminophen

• ​What are phonemes?

• ​Custom pronunciation without needing to learn the international phonetic alphabet (IPA) or brute force spellings

• ​Scaling challenges

• ​Switching to a different TTS provider while in production

• ​Tradeoffs between emotional range and pronunciation consistency

• ​Misleading latency promises and tactics to reduce latency

• ​New tools to catch pronunciation failures before your customers notice

• ​How to set up useful metrics that matter beyond TTFB

• ​War stories from the Voice AI trenches

​If you're actively building and scaling Voice AI applications, this event is for you!

​Agenda:

• ​6:30 – 7:30: Lightning Talks

  • ​Rime: [To be announced]

  • ​ConverseNow: [To be announced]

  • ​Coval: Newest discoveries of benchmarking over 12 different TTS providers & what you should know before picking yours!

  • ​Daily: [To be announced]

• ​7:00 – 7:30: Panel Discussion on how to scale with reliable TTS performance:

• ​7:30 - 8:30: Networking

​About the Hosts

​Rime builds speech AI infrastructure and blends advanced ML with deep linguistic and sociolinguistic insight to create ultra-realistic multilingual voices that breathe, laugh, and carry the subtle rhythms of real everyday speech.

​Coval builds the reliability infrastructure for Voice AI agents, where QA, product, and engineering teams collaborate to test, evaluate, and monitor voice agents at scale.

​ConverseNow builds voice AI for restaurants. ConverseNow delivers a branded and fully customizable experience for each guest to improve customer satisfaction, drive frequency of spend, and give restaurants control through AI Your Way.

Daily is the maintainer of Pipecat, an open source voice orchestration framework for building real-time conversational AI. Pipecat provides the low-latency infrastructure to connect speech recognition, LLMs, and text-to-speech into seamless pipelines, enabling developers to create natural, responsive, and reliable voice experiences.