ConversationalAI4J

A Java library for voice-enabled conversational AI. Speak to an LLM through your browser and get intelligent spoken responses.

Status

This project is in early alpha stage. Features are functional but may contain bugs and APIs may change.

This project was vibe coded, serving as a PoC for PhoneBlock -> "Enhance PhoneBlock-AB with Local AI for Intelligent Scam Call Conversations"

Quick Start

git clone https://github.com/SchulteDev/ConversationalAI4J.git
cd ConversationalAI4J

# Complete voice AI system
# First start NEEDS A FEW MINUTES to download the models
docker-compose up --build
# → http://localhost:8080

Library Usage

// Text chat
ConversationalAI ai = ConversationalAI.builder()
    .withOllamaModel("llama3.2:3b")
    .build();
String response = ai.chat("Hello!");

// Voice chat
ConversationalAI voiceAI = ConversationalAI.builder()
  .withOllamaModel("llama3.2:3b")
  .withSpeech()
  .build();
byte[] audioResponse = voiceAI.voiceChat(audioBytes);

Voice Demo

1. Click microphone button to start recording
2. Speak your message
3. Click microphone again to stop and send
4. AI responds with both text and synthesized speech

Pipeline: Browser Audio → FFmpeg Decoding → Whisper.cpp → Ollama → Piper TTS

Browser Compatibility: Works with modern browsers (Chrome, Firefox, Safari) using MediaRecorder API. Supports WebM/Opus and WAV formats with server-side FFmpeg decoding.

Configuration

Variable	Default	Purpose
OLLAMA_BASE_URL	http://localhost:11434	Ollama server
SPEECH_ENABLED	false	Enable voice features
WHISPER_MODEL_PATH	/app/models/whisper/ggml-base.en.bin	Whisper STT model path
PIPER_MODEL_PATH	/app/models/piper/en_US-amy-medium.onnx	Piper TTS model path
PIPER_CONFIG_PATH	/app/models/piper/en_US-amy-medium.onnx.json	Piper TTS config path

See CONTRIBUTING.md for development setup and ARCHITECTURE.md for technical details.