Voice AI

A helpful template for building voice assistant apps with Groq and cool & creative UI. Built with expo-audio, react-native-reanimated, and Groq APIs.

Project structure

Where things live

All details included in the app readme with deeper details.

voice-assistant/ in detail

• VoiceAssistant.tsx — Main component. Flow states: idle → starting → listening → post_recording → waiting_for_ai → ai_responding. Uses hooks from hooks/, constants from constants/, and formatTime from utils/voice-recording.
• Visuals: IdleFourCircles, ListeningMetering, ThoughtBubbleAnimation, RespondingMeteringCircle.

Groq integration

Three Groq APIs are used; all share one API key (EXPO_PUBLIC_GROQ_API_KEY).

1. Transcription (Speech-to-Text)

• Model: whisper-large-v3-turbo (default)
• Flow: After recording, the audio URI is sent via transcribe(uri) → FormData with file and model; response is verbose_json. The returned text is passed to chat.

2. Chat (LLM reply)

• Model: llama-3.1-8b-instant (default)
• Flow: Transcribed text is sent with complete(userMessage). System prompt asks for short, conversational replies (1–3 sentences for voice). max_tokens: 150, temperature: 0.7. The returned text is passed to TTS.

3. Text-to-Speech (TTS)

• Model: canopylabs/orpheus-v1-english (default), voice hannah
• Flow: Chat reply text is sent with speak(text) → response is binary (WAV). It’s base64-encoded and written to cache with expo-file-system; the file path is passed to the expo-audio player.

Setup and run

1. Install dependencies:

   bash

   npm install

2. Groq API key:

   • Get a key from console.groq.com.
   • Copy .env.example to .env and set the key:
   bash

   cp .env.example .env

   In .env:

   EXPO_PUBLIC_GROQ_API_KEY=your_key_here

3. Start the dev server (restart after changing .env):

   bash

   npx expo start

A detailed readme included in the zip file.