Video Audio Enhancer with Azure OpenAI

This project enhances the audio quality of videos by extracting the audio, converting it into a transcript, correcting grammar, and eliminating filler words using Azure OpenAI. The modified transcript is then converted back into audio and precisely mapped to the original video, ensuring seamless synchronization.

Table of Contents

• Features
• Installation
• Usage
• Project Workflow
• Dependencies
• License
• Flow

Features

• Automatic Speech-to-Text: Extracts audio from the input video and converts it into text using a speech recognition engine.
• Grammar Correction: Corrects grammatical errors in the transcript using Azure OpenAI.
• Filler Word Removal: Removes common filler words such as "uh", "um", and "hmm" from the transcript to improve clarity.
• Text-to-Speech: Converts the cleaned transcript back into audio.
• Seamless Audio-Video Synchronization: Ensures the new audio is perfectly synchronized with the original video, without any delay or mismatch.

Flow

• https://skstanwar.github.io/Curious-PM-/

Installation

1. Clone the repository:

   git clone https://github.com/skstanwar/Curious-PM-.git
   cd Curious-PM-

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # For Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set up Azure OpenAI API:

   • Create an account and get your API key from Azure OpenAI.
   • Set up your API key in the environment file .env:

   azure_openai_key="**********"

Usage

1. Provide your input video: Place the video file in the input directory or specify the path in the script.

2. Run the script:

   python main.py input_video_path.mp4

   • wait for 20 to 30 secs

Project Workflow

1. Audio Extraction: The script extracts the audio track from the input video using MoviePy and saves it as a separate audio file.

2. Speech-to-Text Conversion: The extracted audio is processed using a speech recognition engine to convert it into a transcript. This step generates a text version of the spoken content.

3. Grammar Correction and Filler Word Removal: The transcript is sent to Azure OpenAI, where grammatical errors are corrected, and filler words such as "umm", "uh", and "hmm" are removed for a more professional-sounding transcript.

4. Text-to-Speech Conversion: The cleaned transcript is converted back into an audio file using a text-to-speech engine.

5. Remapping Audio to Video: The newly generated audio is remapped back to the original video. The script ensures perfect synchronization between the new audio and the video, with no delays or mismatches.

Dependencies

• Python: 3.8+
• MoviePy: For video processing
• Azure OpenAI: For interacting with Azure OpenAI API
• deepgram: For converting audio to text and text to audio
• streamlit: For web app view application

License

This project is licensed under the MIT License. See the LICENSE file for more details.