Skip to main content
Pocket TTS is a lightweight Text-to-Speech model from Kyutai that runs efficiently on CPU. It offers low latency (~200ms to first audio), voice cloning support, and multiple built-in voices, making it ideal for local voice synthesis without requiring a GPU or external API. The Pocket TTS plugin for the Stream Python AI SDK allows you to add local TTS functionality to your project.

​
Installation

Install the Stream Pocket TTS plugin with 
uv add vision-agents[pocket]

​
Example

Check out our Pocket TTS example to see a working code sample using the plugin, or read on for some key details.

​
Initialisation

The Pocket TTS plugin exists in the form of the TTS class: 
from vision_agents.plugins import pocket

# Create TTS with default voice
tts = pocket.TTS()

# Or specify a built-in voice
tts = pocket.TTS(voice="marius")

# Or use a custom voice for cloning
tts = pocket.TTS(voice="path/to/your/voice.wav")

​
Parameters

These are the parameters available in the Pocket TTS plugin for you to customise:
NameTypeDefaultDescription
voicestr"alba"Built-in voice name or path to a custom wav file for voice cloning. Built-in options: alba, marius, javert, jean, fantine, cosette, eponine, azelma.
clientTTSModel or NoneNoneOptional pre-initialized TTSModel instance for advanced use cases.

​
Built-in voices

Pocket TTS includes several built-in voices:
  • alba - Default voice
  • marius
  • javert
  • jean
  • fantine
  • cosette
  • eponine
  • azelma

​
Voice cloning

You can clone a voice by providing a path to a wav file: 
from vision_agents.plugins import pocket

# Use a local wav file
tts = pocket.TTS(voice="path/to/your/voice.wav")

# Or use a HuggingFace-hosted voice
tts = pocket.TTS(voice="hf://kyutai/tts-voices/alba-mackenna/casual.wav")

​
Features

  • CPU-only - No GPU required, runs efficiently on standard hardware
  • Low latency - ~200ms to first audio
  • Small model size - 100M parameters
  • Voice cloning - Use custom wav files for voice cloning
  • Built-in voices - 8 pre-configured voices available