How to stream text to speech in Python over a WebSocket

AssemblyAI’s Python text-to-speech tutorial explains how to progress from generating a one-off WAV file through an HTTP request to building a production-oriented streaming voice agent over a persistent WebSocket session. It presents the AssemblyAI-TTS-1 model, which supports six languages, configurable voices, raw PCM streaming, telephony formats, and character-based pricing, while emphasizing explicit voice and language settings and reading audio configuration from server responses. The tutorial demonstrates saving streamed PCM with a WAV header, playing it live through sounddevice, and connecting TTS to a streaming LLM by sending tokens incrementally, flushing at sentence boundaries, and retrieving audio concurrently. A central focus is barge-in handling: when a caller interrupts, the application should halt LLM generation, cancel queued and in-flight server synthesis, and clear locally buffered playback while keeping the session open for the next turn. It also covers session reliability, including keep-alive messages for two-minute idle timeouts, one-hour session lifetimes, text and control-frame limits, and manual raw-WebSocket protocol handling. The broader recommendation is to treat TTS as part of a coordinated real-time STT-LLM-TTS loop, where session reuse, low time to first audio, sentence-level flushing, interruption behavior, and connection management determine how natural a voice agent feels.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.