Skip to main content

​
Overview

VoiceAiTTSService converts text to speech using Voice.ai’s Multi-Context WebSocket API. It maintains a persistent WebSocket connection that streams raw PCM audio (32kHz mono), with support for multiple languages, custom voice selection, and temperature/top_p generation controls.

Source Repository

Source code, examples, and issues for the Voice.ai integration

Voice.ai Website

Sign up and get a Voice.ai API key

API Documentation

Voice.ai Multi-Context WebSocket TTS API reference

​
Installation

This is a community-maintained package distributed separately from pipecat-ai. It is not published to PyPI; install it from source: 
git clone https://github.com/voice-ai/voice-ai-pipecat-tts.git
cd voice-ai-pipecat-tts
pip install -e .

​
Prerequisites

​
Voice.ai Account Setup

Before using the Voice.ai TTS service, you need:
  1. Voice.ai Account: Sign up at Voice.ai
  2. API Key: Obtain an API key (format: vk_*) from Voice.ai

​
Required Environment Variables

  • VOICEAI_API_KEY: Your Voice.ai API key for authentication

​
Configuration

​
api_key
str
required
Voice.ai API key for authentication (format: vk_*).
​
voice_id
str
default:"None"
Voice identifier for synthesis. If not provided, uses the default built-in voice.
​
url
str
default:"wss://dev.voice.ai/api/v1/tts/multi-stream"
WebSocket URL for the Voice.ai multi-context TTS API.
​
sample_rate
int
default:"None"
Output audio sample rate. Defaults to 32000 Hz (Voice.ai’s native rate) when not set.
​
params
VoiceAiTTSService.InputParams
default:"None"
Voice synthesis settings. See Input Parameters below.
​
aggregate_sentences
bool
default:"True"
Whether to aggregate text by sentences before TTS. When True, each sentence is sent separately for lower latency; when False, larger text chunks are batched for more natural flow at the cost of higher latency.

​
Input Parameters

Synthesis settings passed via the params constructor argument using VoiceAiTTSService.InputParams(...).
ParameterTypeDefaultDescription
languageLanguageLanguage.ENTarget language. Supports en, ca, sv, es, fr, de, it, pt, pl, ru, nl.
modelstrNoneTTS model. Auto-selected based on language when not provided.
audio_formatstr"pcm"Audio output format (raw PCM).
temperaturefloat1.0Generation temperature (0.0–2.0). Higher values are more random.
top_pfloat0.8Top-p sampling (0.0–1.0). Controls output diversity.
Available parameters and defaults are defined by the integration and the Voice.ai API. See the source repository for the authoritative, up-to-date list.

​
Usage

from pipecat_voice_ai import VoiceAiTTSService
from pipecat.transcriptions.language import Language

tts = VoiceAiTTSService(
    api_key="vk_your-api-key",
    voice_id="your-voice-id",  # Optional: uses default if not provided
    params=VoiceAiTTSService.InputParams(
        language=Language.EN,
        temperature=1.0,
        top_p=0.8,
    ),
)

# Add tts to your pipeline:
from pipecat.pipeline.pipeline import Pipeline

pipeline = Pipeline([
    # ... other processors ...
    tts,
    transport.output(),
])

​
Compatibility

Tested with Pipecat v0.0.100+. Check the source repository for the latest tested version and changelog.