You're cooking dinner and your hands are covered in flour. You need to know how long to roast the chicken at 375°. You could wash your hands, dry them, unlock your phone, type the question... or you could just say "Hey Meggy, how long do I roast a chicken at 375?" and get an answer spoken back to you.
Voice Chat turns Meggy into a hands-free assistant. It listens for your voice, understands what you say, thinks about the answer using the same powerful AI engine behind text conversations, and speaks the response back to you. Same tools, same memory, same intelligence — just no keyboard required.
Voice Chat is built on a four-stage pipeline:
Meggy listens for its wake word — "Hey Meggy" — using a lightweight local detection model. This runs continuously in the background without sending any audio to the cloud. When the wake word is detected, the microphone activates and recording begins.
You can also use push-to-talk mode if you prefer — hold a key to speak, release to send. Both modes are available in settings.
Once you finish speaking, your audio is transcribed into text. Meggy supports multiple STT providers:
| Provider | Model | Runs Locally? |
|---|---|---|
| Local Whisper | Whisper.cpp | ✅ Yes — fully on-device |
| OpenAI | Whisper | Cloud |
| Deepgram | Nova-3 | Cloud |
| Groq | Whisper Large v3 Turbo | Cloud — extremely fast |
| AssemblyAI | Universal-2 | Cloud — high accuracy |
If privacy is your priority, the local Whisper option means your voice never leaves your machine. If speed matters most, Groq’s LPU-accelerated Whisper is the fastest cloud option.
The transcribed text is processed through the exact same AI pipeline as typed messages. This means Voice Chat has access to:
You can ask voice questions that trigger tool calls — "What's the weather like?", "Turn off the bedroom lights", "Add milk to my shopping list" — and Meggy will use the appropriate tools to fulfill the request.
The AI's response is spoken back to you using natural-sounding voice synthesis:
| Provider | Voices | Quality |
|---|---|---|
| ElevenLabs | Hundreds of natural voices | Premium, highly expressive |
| OpenAI | 6 built-in voices (Alloy, Echo, Fable, Onyx, Nova, Shimmer) | High quality, fast |
| Deepgram | Aura-2 voices | Low latency, natural |
| Cartesia | Sonic-2 voices | Ultra-low latency, expressive |
| Piper | Local voices | ✅ Runs offline — no cloud needed |
| Web Speech API | Browser built-in voices | Zero setup — works everywhere |
You can choose your preferred voice in settings, adjusting speed, pitch, and provider.
Meggy uses Voice Activity Detection to know when you've finished speaking. VAD analyzes the audio stream in real time to detect speech boundaries — it knows when you start talking and when you stop, so it doesn't cut you off mid-sentence or wait awkwardly after you've finished.
Voice Chat works on all supported platforms:
Want to interrupt Meggy mid-sentence? With conversational mode enabled, you can barge in at any time — correct the AI, change the topic, or redirect the conversation — and Meggy will stop, understand what you meant, and pivot instantly.
An echo gate prevents false triggers by distinguishing your voice from the AI's own speech playing through the speakers. During TTS playback, only audio significantly louder than the speaker baseline passes through to the speech recognizer. After playback ends, a brief cooldown blocks residual echo.
You can tune interrupt sensitivity from 1 (very conservative — only explicit corrections) to 5 (very aggressive — almost any speech interrupts). The default level of 3 works well for most conversations.
To enable it, go to your agent's channel bindings and toggle Conversational Mode on for the voice channel. See the Conversational Agents article for the full deep-dive.
For the ultimate low-latency experience, Meggy supports Live Audio Mode. Instead of the traditional multi-stage pipeline (Voice → STT → LLM → TTS → Voice), Live Audio uses a direct bidirectional speech-to-speech connection.
This eliminates round-trip delays, offering sub-500ms latency that feels like talking to a real human. It supports native conversational barge-in, meaning you can interrupt the AI organically and it will stop instantly, without needing the traditional echo gate.
Supported Live Audio providers include:
You can toggle Live Audio mode on directly from the Voice & Speech settings. When enabled, it short-circuits the traditional pipeline for unprecedented speed.
Tip: You can also configure Voice & Speech during the onboarding wizard when you first set up Meggy.
Voice Chat integrates with all of Meggy's channels — you can start a voice conversation on desktop and continue it via text on WhatsApp, or vice versa. It's all the same conversation, the same memory, the same assistant.