Sign in Join

Live chat for Free Pascal, Lazarus, Delphi and every Pascal programmer.

Catalog · How the catalog works
Anything · Lazarus or Delphi
Random projectRandom project

kai-google-voice-sample

View on GitHub

Member rating

No ratings yet.

Sign in to rate this project.

README

Gemini Voice Live with Embarcadero Delphi & Node.js

A full-duplex real-time voice and simultaneous translation application built using Embarcadero Delphi 13 (FMX / WinRT / Windows Multimedia PCM) and a Node.js / Express WebSocket Backend powered by Google's Gemini Multimodal Live API (gemini-3.1-flash-live-preview).

Note: This entire sample was elaborated with the help of Embarcadero's official Kai code agent.

🌟 Overview & Features

  • Full-Duplex Interactive Voice Assistant: Real-time conversational AI in Brazilian Portuguese with ultra-low latency.
  • Simultaneous Translator (English): Live speech-to-English translation streaming both synthesized audio and transcribed text in real time.
  • Native Delphi Audio Engine (Audio.PCM.Windows.pas): Direct low-latency PCM audio capture (16kHz, 16-bit Mono) with Voice Activity Detection (RMS) and PCM playback (24kHz, 16-bit Mono) using Windows Multimedia (winmm.dll).
  • Async WebSocket Client (Gemini.Live.Client.pas): Asynchronous threaded communication with the backend relay using Windows WinRT IMessageWebSocket.
  • Intelligent Echo & Interruption Handling: Prevents acoustic feedback during AI speech and seamlessly handles model interruptions.
  • Elaborated with Embarcadero Kai: Architecture, audio driver integration, WinRT asynchronous streaming, and backend relay designed with Embarcadero's official Kai coding agent.

🏗 Architecture

pascal
??????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????
???              Embarcadero Delphi Client                 ???
???                     (Delphi 13)                        ???
???                                                        ???
???  ???????????????????????????????????????????????????????????????????????????   ????????????????????????????????????????????????????????????????????????  ???
???  ???   TWindowsPcmAudio    ???   ???  TGeminiLiveClient   ???  ???
???  ???  - 16kHz PCM In (Mic) ???   ???  - WinRT WebSocket   ???  ???
???  ???  - 24kHz PCM Out (Spk)???   ???  - Message Queues    ???  ???
???  ???  - RMS VAD & Echo Mute???   ???  - State Management  ???  ???
???  ???????????????????????????????????????????????????????????????????????????   ????????????????????????????????????????????????????????????????????????  ???
??????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????
                            ??? WebSocket (JSON + Base64 PCM)
                            ??? ws://127.0.0.1:3000/live
??????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????
???                  Node.js Backend Relay                 ???
???                                                        ???
???  - Express HTTP & WebSocket Server (`server.ts`)       ???
???  - Google GenAI SDK (`@google/genai`)                  ???
???  - Model: `gemini-3.1-flash-live-preview`              ???
??????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????
                            ??? Gemini Live Bidirectional Stream
                            ???
               Google Gemini Live API

🤖 Developed with Embarcadero Kai

This project was elaborated and accelerated using Embarcadero's official Kai code agent, demonstrating seamless integration between modern Delphi 13 FireMonkey applications, Windows native audio multimedia APIs (winmm), WinRT asynchronous WebSockets, and Google's Gemini Multimodal Live real-time audio protocols.

📁 Repository Structure

text
kai-google-voice-sample/
├── delphi/                     # Delphi 13 FMX Client Application
│   ├── DelphiVoiceSample.dpr   # Delphi project source file
│   ├── DelphiVoiceSample.dproj # Delphi project configuration
│   ├── DelphiVoiceSample.res   # Delphi project resource
│   ├── Form.Main.pas / .fmx    # Main UI form and interaction handlers
│   ├── Gemini.Live.Client.pas  # Asynchronous WinRT WebSocket client
│   └── Audio.PCM.Windows.pas   # Native Windows PCM WaveIn/WaveOut driver
├── backend/                    # Node.js WebSocket Voice Relay
│   ├── server.ts               # Express & WebSocket Gemini Live proxy
│   ├── package.json            # Node.js dependencies and scripts
│   ├── .env.example            # Environment variable template (NEVER commit .env)
│   ├── README.md               # Backend documentation
│   └── src/                    # Web frontend components
├── .env.example                # Root environment template
├── .gitignore                  # Excludes binaries, node_modules, and secrets
├── LICENSE                     # License file
└── README.md                   # Project documentation

🚀 Getting Started

Prerequisites

  1. Embarcadero Delphi 13 (or compatible) with FireMonkey and Windows SDK.
  2. Node.js (v18+) and npm.
  3. Google Gemini API Key from Google AI Studio.

Step 1: Start the Backend Relay

  1. Navigate to the backend folder:

    bash
    cd backend

  2. Install dependencies:

    bash
    npm install

  3. Create your .env file from .env.example:

    bash
    cp .env.example .env

  4. Open .env and set your GEMINI_API_KEY:

    ini
    GEMINI_API_KEY=your_actual_gemini_api_key_here
PORT=3000

  5. Run the backend development server:

    bash
    npm run dev

    The backend server will start on http://127.0.0.1:3000 and listen for Delphi WebSocket connections on ws://127.0.0.1:3000/live.

Step 2: Run the Delphi Client

  1. Open delphi/DelphiVoiceSample.dproj in Embarcadero RAD Studio / Delphi 13.
  2. Select the target platform (Win32 or Win64).
  3. Build and run the application (F9).
  4. Click:
    • "Iniciar Gemini Live" to start an interactive conversation in Brazilian Portuguese.
    • "Iniciar Tradutor Inglês" to translate your spoken Portuguese to English with real-time voice and transcript text.

🔒 Security Notes

  • Never commit your .env file or API keys to GitHub.
  • The .gitignore in this repository is pre-configured to ignore .env* files, build binaries (*.dcu, *.exe, Win32/, Win64/), and node_modules/.