Overview

Blynt builds a speech recognition engine tuned for your product and runs it as a dedicated deployment. This documentation shows how to connect to that deployment and transcribe audio in real time over a WebSocket, so you can run your custom model in production with a few lines of client code.

Key features

• Real-time streaming transcription: Stream audio over a WebSocket and receive incremental turn_partial and final turn_ended transcripts as the user speaks. See Streaming audio.
• Client-controlled turns: You explicitly mark where each turn starts and ends with start_turn and end_turn, giving you full control over when the engine listens. See Turn-taking mode.
• Session- and turn-level contextual biasing: Boost recognition accuracy by declaring values you expect the user to say. Session-level context on start_session applies to every turn; turn-level turnContext on start_turn is added for that turn. Most effective for short, constrained answers picked from a known list (niche vocabulary, brand names, or fixed option sets), typical of elicitation prompts.

Learn more