TTS Modules

February 18, 2026 · View on GitHub

← Back to README | Architecture

Overview

The TTS layer is defined by TTSInterface. All engines generate a NumPy audio array + sample rate, which the brain plays via sounddevice. The active engine is selected with tts_provider in config.

src/modules/tts/
├── edge_tts_wrapper.py    Microsoft EdgeTTS (free, online)
├── kokoro_tts_wrapper.py  Kokoro ONNX (local, no API)
└── orpheus_tts_wrapper.py Orpheus (API, high quality)

Interface Contract

class TTSInterface(ABC):
    async def generate_audio(text: str) -> (np.ndarray, sample_rate: int)
    async def speak(text: str, output_device_id: int) -> None
    def reload_config(config: BrainConfig) -> None

The brain always calls generate_audio() and handles playback itself via sounddevice.play(). speak() is also an abstract method — implementations must provide it (even if only as a thin wrapper around generate_audio()). The Kokoro wrapper includes a working speak() for direct use; the brain itself does not call it.

Important for custom TTS engines: if you omit speak() from your implementation, Python will raise TypeError at instantiation time because it is declared @abstractmethod in TTSInterface. This allows interrupt/resume functionality (the audio buffer is tracked at the brain level).

Providers

EdgeTTS (edge_tts_wrapper.py)

Library: edge-tts
Cost: Free (uses Microsoft Edge's TTS API)
Config keys: tts_voice, tts_pitch, tts_rate, tts_volume

Generates audio to a temporary MP3 file, reads it back as a NumPy array via soundfile, then deletes the temp file. Each generation uses a unique UUID filename to avoid collisions during concurrent calls.

Voice format: "it-IT-IsabellaNeural", "en-US-AvaNeural", etc.
Full voice list: edge-tts --list-voices

tts = EdgeTTSWrapper(voice="en-US-AvaNeural", pitch="+5Hz", rate="+10%", volume="+33%")
audio, sr = await tts.generate_audio("Hello!")

Constructor note: EdgeTTSWrapper.__init__ also accepts an output_file parameter (default: "temp_tts.mp3"). This parameter is vestigial — generate_audio() ignores it and always uses a unique UUID-based filename to prevent collisions during concurrent calls. It is safe to omit. The class-level default for voice is "en-US-JennyNeural"; at runtime the value from BrainConfig.tts_voice ("en-US-AvaNeural") is always passed explicitly.

Default mismatch note (EdgeTTS): The class-level constructor defaults for pitch, rate, and volume are "+0Hz", "+0%", "+0%" respectively — these differ from the BrainConfig defaults of "+5Hz", "+10%", "+33%". The brain always passes the config values explicitly, so the class defaults only matter if EdgeTTSWrapper is instantiated directly without arguments (e.g. in tests or standalone usage).

Dead method note: EdgeTTSWrapper contains a vestigial generate_audio(self, text: str, filename: str) -> None definition (the original helper that wrote to a fixed filename). Python silently shadows it with the second generate_audio(self, text: str) -> tuple[np.ndarray, int] definition, which is the one that actually executes. The first definition is unreachable and has no effect at runtime.

Kokoro ONNX (kokoro_tts_wrapper.py)

Library: kokoro-onnx
Cost: Free (runs entirely locally)
Config keys: kokoro_model, kokoro_voices_file, kokoro_voice, kokoro_speed, kokoro_lang

Runs the Kokoro TTS model locally via ONNX Runtime. No internet connection required after downloading the model files. Best for privacy or offline use.

Model files: kokoro-v0_19.onnx and voices.bin are downloaded automatically on first launch if missing (from GitHub Releases, ~125 MB total). No manual download needed.

To use a custom path, update kokoro_model and kokoro_voices_file in config.json.

Voice examples: af_bella, af_sarah, am_adam, bf_emma

Orpheus (orpheus_tts_wrapper.py)

Library: requests
Cost: API-based (Baseten — billed per inference)
Config keys: orpheus_voice
Env vars (secrets — never saved to config.json): ORPHEUS_API_KEY, ORPHEUS_ENDPOINT

Calls a self-deployed Orpheus model on Baseten. Produces highly expressive, human-like speech — the highest quality TTS option available.

Setup required: You must deploy the Orpheus model to your own Baseten workspace before use. See Setup Guide → Orpheus TTS Setup for step-by-step instructions.

The wrapper POSTs to your endpoint with stream: true, collects raw PCM bytes (24 kHz, 16-bit mono), decodes them to a NumPy array, and returns them to the brain for playback.

Voice examples: zoe, tara, leo, leah

Default mismatch note (Orpheus): The OrpheusTTSWrapper class constructor defaults to voice="tara". The BrainConfig dataclass default for orpheus_voice is "zoe". At runtime the brain always passes config.orpheus_voice explicitly, so the effective default seen by users is "zoe". The class default only matters for direct instantiation without arguments.

Hot Reload

reload_config() updates voice in place for EdgeTTS (also pitch, rate, volume). For Kokoro it updates voice, speed, and lang. For Orpheus, it also updates the API key, endpoint URL, and voice. Changing tts_provider itself requires a restart (the object type changes).

Adding a New TTS Engine

  1. Create src/modules/tts/my_tts.py and extend TTSInterface.
  2. Implement async generate_audio(text) -> (np.ndarray, int) and reload_config().
  3. In main.py, add the instantiation branch.
  4. Add the provider name to --tts-provider choices.