Stream Recorder and Transcriber GUI

**Real-time audio stream recorder and speech-to-text transcriber** powered by [faster-whisper](https://github.com/SYSTRAN/faster-whisper). This desktop application captures live audio streams (e.g., news radio, online broadcasts) via FFmpeg, transcribes speech in real time using a local Whisper model, and logs both audio and text output.

Note: This is a self-contained PySide6 GUI application designed for Windows (with ffmpeg.exe support), but can be adapted to other platforms.

---

✨ Features

• Real-time transcription of live audio streams (HLS, Icecast, HTTP, etc.)
• Predefined stream presets (BBC, CNN, Al Jazeera, VOA, Bloomberg, and more)
• Manual URL input for custom streams
• Configurable recording duration (1–30 minutes)
• Adjustable audio chunk size for latency/performance tuning
• Live transcription display with color-coded formatting
• Automatic audio saving as WAV → MP3 (192 kbps)
• Usage logging with stream name, action type, and duration
• Model-on-demand loading: Whisper base model (int8, CPU-only) loads only when needed
• Listen-only mode: transcribe without saving audio

---

🧠 Model Details

• Uses faster-whisper with the base model (English-optimized)
• Runs locally on CPU (int8 quantization for efficiency)
• Language is locked to English (with VAD filtering enabled)
• Model is loaded once per session and reused across recordings

---

📦 Requirements

• Python 3.8+
• Required packages:

  pip install faster-whisper pyside6 pyaudio pydub colorama requests numpy

• FFmpeg (bundled as ffmpeg.exe in frozen builds; otherwise must be in PATH)
• ~500 MB disk space for model cache (~/.cache/huggingface/hub)

---

🚀 Quick Start

1. Clone or download the project.
2. Install dependencies:

   pip install -r requirements.txt  # if you create one

3. Run the application:

   python stream.py

4. Choose a stream from the dropdown or enter a custom URL.
5. Set duration (1–30 min), optionally adjust chunk size, and click Record.

The first run will download the Whisper base model (~150 MB).

---

📁 Output

• Text: displayed in real time; each new sentence appears in red bold, then turns blue bold after processing.
• Audio: saved as StreamName_audio_YYYY-MM-DD_HH-MM-SS.wav → auto-converted to .mp3.
• Log: stream.log in the app directory with entries like:
  2025-12-11 14:30:22 BBC - Record: 2:15

---

🛠️ Configuration Options (UI)

Control	Purpose
Stream selector	Choose from built-in stations or type a URL
Duration (min)	Max recording/transcription time
Audio chunk size	Controls block duration: 160000 = 5s, 320000 = 10s, ..., 1920000 = 60s
Record button	Start/stop recording + transcription + audio save
Listen button	Transcribe only (no file saved)

---

📜 License

This project is for personal/educational use.
Underlying components:

• faster-whisper — MIT License
• FFmpeg — GPL/LGPL
• PySide6 — [LGPL/GPL]

---

💡 Notes

• Designed for stable, low-bandwidth English streams (news, talk radio).
• Not optimized for noisy or multilingual content.
• Avoid very short chunk sizes (<5s) — may reduce transcription accuracy.
• On first launch, ensure internet access for model download and stream validation.

---

Made with ❤️ using Python, faster-whisper, and PySide6.
Version: V211125

---