Streaming Text-to-Speech API Documentation

WebSocket Connection

wss://api.voxentis.com/v1/tts/stream?voice_id=pNInz6obpg&format=pcm16

Query Parameters

Sending Text

{"type": "text", "content": "Hello! "}
{"type": "text", "content": "How can I "}
{"type": "text", "content": "help you today?"}

// Signal end of text to flush remaining audio
{"type": "flush"}

// Close the connection
{"type": "close"}

Python — Stream TTS from LLM

import asyncio
import websockets
import json

async def stream_tts(llm_tokens):
    uri = "wss://api.voxentis.com/v1/tts/stream?voice_id=pNInz6obpg"
    headers = {"Authorization": "Bearer YOUR_API_KEY"}

    async with websockets.connect(uri, extra_headers=headers) as ws:
        async def receive_audio():
            async for message in ws:
                if isinstance(message, bytes):
                    play_audio(message)

        receiver = asyncio.create_task(receive_audio())

        for token in llm_tokens:
            await ws.send(json.dumps({
                "type": "text",
                "content": token
            }))

        await ws.send(json.dumps({"type": "flush"}))
        await asyncio.sleep(1)
        await ws.send(json.dumps({"type": "close"}))
        await receiver

Ultra-Low Latency: Streaming TTS produces the first audio frame within ~200ms of receiving text.

Handling Interruptions

If the caller interrupts while the agent is speaking, send a cancel message to stop audio generation immediately:

{"type": "cancel"}

When used within a Voxentis voice agent, streaming TTS and interruption handling are managed automatically by the pipeline.