Скачать или смотреть Building the Next Generation of Voice AI

This source, an *excerpt from a YouTube video transcript**, features **Toki Sherbakov and Anoop Kotha from OpenAI* discussing the *advancements and practical applications of audio-based AI agents**. They highlight the **shift from text-centric AI to a multimodal era**, emphasizing the **improved speed, expressiveness, and accuracy of real-time speech-to-speech models* compared to older, less efficient chained architectures. The speakers *outline key considerations for building these agents**, such as **latency, cost, accuracy, user experience, and integrations**, while also **exploring patterns like delegation through tools and the importance of voice customization through prompting**. Finally, they touch upon **evaluation methods and guardrails for ensuring safety and performance* in these emerging AI solutions.