Video-To-Text Scripts

This repository contains a collection of Python scripts with a graphical user interface (PyQt6) designed for a full cycle of video processing: from audio extraction to text summarization.

� Features

1. Audio Extractor (extractor.py)

A utility for extracting audio tracks from video files. Prepares audio for further recognition.

• Input Formats: .mp4, .mkv, .avi, .mov, .wmv
• Output Formats:
  • .wav (PCM 16kHz Mono) — Recommended format for speech recognition.
  • .mp3 — Compressed format to save space.
• Key Features: Automatic conversion of sampling rate and channels.

2. Speech Recognizer (recognizer.py)

A tool for automatic speech recognition (Speech-to-Text) based on the Whisper model.

• Engine: faster-whisper (optimized version of OpenAI Whisper).
• Operation Mode: Local execution on CPU (with int8 quantization to reduce memory consumption).
• Input Formats: .wav, .mp3, .m4a, .flac.
• Output: A .txt file containing the full recognized text.
• Note: Currently configured to recognize Russian speech (language="ru").
• Default model: medium (good quality for 4 CPU / 8 GB RAM server profile).

3. Summarizer (summarizer.py)

A smart text summarizer using the YandexGPT API.

• Functionality: Creates a summary of large texts, lectures, or interviews.
• Key Features:
  • Automatic slicing of long texts into chunks of 8000 characters.
  • Support for custom system prompts (instructions for the neural network).
  • Operates via the official Yandex Cloud API.
• Requirements: You must provide an API Key, Folder ID, and Model URI.

---

� Installation and Usage

Prerequisites

To run the scripts, you need Python 3.8+ installed.

Installing Dependencies

Run the following command in your terminal:

pip install PyQt6 moviepy faster-whisper requests

Note: If you encounter errors with moviepy, ensure you have a compatible version installed or check the script's built-in exception handling.

How to Run

Execute the scripts sequentially depending on your task:

1. Extract audio from video:

   python extractor.py

2. Recognize text from audio:

   python recognizer.py

3. Generate a summary:

   python summarizer.py

---

📝 Notes

• To use summarizer.py, you need a Yandex Cloud account with a service account that has the ai.languageModels.user role.
• The selected Whisper model will be downloaded automatically on first run. This may take some time.
• You can tune recognizer via environment variables:

WHISPER_MODEL_SIZE=medium
WHISPER_COMPUTE_TYPE=int8
WHISPER_CPU_THREADS=4
WHISPER_NUM_WORKERS=1
WHISPER_BEAM_SIZE=5
WHISPER_LANGUAGE=ru
WHISPER_VAD_FILTER=1
python recognizer.py

• If recognition is too slow, switch to WHISPER_MODEL_SIZE=small.