Audio Transcription PoC 🎤 → 📝

A simple Proof of Concept (PoC) application that records audio from the microphone in a web front-end, sends it to a Fastapi API, and transcribes it using OpenAI's Whisper model.

📌 Features

• Record audio directly from the browser
• Send recorded audio to a Fastapi backend
• Use OpenAI Whisper for speech-to-text transcription
• Display the transcribed text in the UI

📦 Tech Stack

• Frontend: JavaScript (Vanilla JS) + HTML + CSS
• Backend: Python + Fastapi
• Transcription Model: OpenAI Whisper
• Audio Processing: ffmpeg

🚀 Setup & Installation

1️⃣ Install Backend Dependencies

Make sure you have Python 3.8+ installed.

1. Clone the repository

git clone https://github.com/abonvalle/PoC-OpenAI-Whisper.git
cd PoC-OpenAI-Whisper

2. Create a virtual environment (optional but recommended)

python -m venv .poc_openai_whisper_env
source .poc_openai_whisper_env/bin/activate # On Windows: .poc_openai_whisper_env\Scripts\activate

3. Install Python dependencies

pip install -r requirements.txt

4. Install ffmpeg (Required for audio processing)

Linux (Ubuntu/Debian):

sudo apt install ffmpeg

macOS:

brew install ffmpeg

Windows (Chocolatey):

choco install ffmpeg

---

2️⃣ Run the Fastapi Backend

1. Run Fastapi server

fastapi dev app.py --port 5000

The API should now be running at: http://localhost:5000

---

3️⃣ Run the Frontend

Use the live server extension https://marketplace.visualstudio.com/items?itemName=ritwickdey.LiveServer (recommended) or open index.html in a browser.

🛠️ API Endpoints

🎙️ Transcribe Audio

Endpoint:

POST /transcribe

Request:

• Content-Type: multipart/form-data
• Body:
  • file: The recorded audio file (WebM/WAV)

Response (JSON):

{
  "message": "Success",
  "transcription": "Hello, this is a test transcription."
}

📜 Project Structure

/PoC-OpenAI-Whisper
│── index.html # Main UI
│── script.js # Handles recording & API calls
│── app.py # Fastapi application
└── requirements.txt # Python dependencies

TODO : update readme with fastapi

fastapi dev app.py --port 5000;