VideoSDK has introduced a new Gladia Speech-to-Text (STT) plugin for its AI Agents SDK. This integration addresses a core challenge in voice agent development: slow or inaccurate transcription, which cripples downstream reasoning. Gladia's STT is built for real-time use, offering low latency and robust multilingual support.

The plugin is optimized for dynamic conversation environments where users might code-switch between languages mid-sentence. It provides partial transcripts, allowing agents to begin processing before a user finishes speaking. This combination of speed and flexibility makes it suitable for live calls, interactive voice applications, and global user bases.

Developers can install the plugin via pip and configure it using a cascading pipeline. Key parameters include language detection lists, code-switching toggles, and audio format settings. By serving as a reliable input layer, Gladia STT aims to ensure that the entire voice agent pipeline remains responsive and accurate, even in complex, multilingual scenarios.