Jacky — Your autonomous desktop companion

Jacky can listen to your voice and talk back! The voice settings panel is divided into two primary sub-tabs to configure Text-to-Speech (TTS) output and Speech-to-Text (STT) input.

Text-to-Speech (TTS)

Configure how Jacky speaks. Jacky supports three distinct Text-to-Speech providers to synthesize audio.

Text-to-Speech (TTS) settings sub-tab

TTS Providers

Provider	Type	Description
Piper	Local	Runs a custom ONNX voice model entirely on your machine — no API key, no internet, zero latency. Point `piper_model_url` to any Piper-compatible `.onnx` model and Jacky downloads it automatically.
ElevenLabs	Cloud	High-quality cloud voices. Requires an API key from elevenlabs.io.
Native	System	Uses the OS built-in speech engine (SAPI on Windows, `say` on macOS). No setup needed.

Configuration Details (tts)

Response Mode (Modo de respuesta): Choose how Jacky replies (e.g. text only, voice only, or both).
Volume (Volumen): Adjust the playback volume of synthesized voices.
TTS Provider Config (tts_provider): The config key accepts "piper", "elevenlabs", "native", or "auto" (tries ElevenLabs → native fallback). Piper and ElevenLabs both automatically fall back to native TTS on failure so Jacky is never silent.
Voice Speed (Velocidad): Adjust speech speed (in words per minute, WPM).
Voice Test (Probar voz): Click to play a test phrase with your current settings.
Piper TTS Custom Model: Point to any Piper .onnx model URL (such as our default chibi voice model) to download and use a completely local custom voice model.
Local Caching: All synthesized audio is cached locally (in ~/.jacky/tts_cache/) so repeated phrases play instantly.

Speech-to-Text (STT)

Configure how Jacky listens to you. Jacky uses voice activity detection (VAD) and speech recognition engines to transcribe your voice.

Speech-to-Text (STT) settings sub-tab

STT Providers

Local (Whisper ONNX) — Completely offline, highly accurate speech recognition using sherpa-onnx Whisper models (tiny, base, or small). Models are downloaded automatically on demand directly to your cache directory.
Whisper (Groq) — Whisper models via Groq API (whisper-large-v3). Uses your Groq API keys if configured.
AssemblyAI — Real-time cloud streaming with a local VAD pre-gate (via webrtcvad / silero) to reduce costs — the billed WebSocket only opens once speech is detected. Requires an API key.

Configuration Details