Real-Time Transcription 📜🎤

Overview

This project is a real-time audio transcription system that captures and transcribes speech using Silero VAD and Faster-Whisper. The system detects voice activity, manages silence, and automatically saves audio segments along with their corresponding transcriptions.

Features

• Real-time Voice Activity Detection: Utilizes Silero VAD to accurately detect when speech is present.
• Automatic Transcription: Leverages the Faster-Whisper model to transcribe audio segments efficiently.
• File Management: Automatically saves recorded audio files and appends transcriptions to a newly created transcript file for easy organization.
• User-Friendly Interface: Simple command-line interface to start recording and manage audio files.

Installation

1. Clone the repository:

   git clone https://github.com/dave111111111/real-time-transcription.git
   cd real-time-transcription
   

2. Install requirements:

   pip install numpy sounddevice torch silero_vad faster_whisper transformers pydub

3. Installing ffmpeg

If you're on Windows, you can easily install FFmpeg using Chocolatey. Follow these steps:

1. Open a command prompt as Administrator.
2. Run the following command:

   choco install ffmpeg
   

Usage

python rec_speak_trans.py

Video Example

Real_time_trasncription.1.mp4

Note on Performance

-For optimal performance, it is recommended to use CUDA for faster processing. Please install CUDA 12.1 and cuDNN 8.9.7 to take advantage of GPU acceleration.

Modifying the Silence Threshold

-To modify the silence threshold for voice activity detection, open the main.py file and adjust the SILENCE_THRESHOLD variable located near the beginning of the file. This threshold determines how sensitive the system is to silence, impacting when the recording starts and stops.

Recording

Once started, the system will:

• Continuously listen to audio input.
• Detect and record speech segments.
• Save each detected segment into the recordings/ directory as a .wav file.

Transcription

The system automatically:

• Starts a transcription thread for each detected segment.
• Saves transcript chunks to the transcripts/ directory, generating a new file if one exists.

Exit & Cleanup

Use Ctrl+C to exit. On exit, all individual audio files are combined into a single output file.

Directory Structure

• recordings/: Stores individual audio segments in .wav format.
• transcripts/: Stores generated transcript files.

Example Output

recordings/ ├── output_sentence_audio_1.wav └── output_sentence_audio_2.wav

transcripts/ └── transcript_1.txt

How It Works

The real-time transcription system utilizes a combination of voice activity detection and automatic transcription to effectively capture and transcribe speech. Here’s a breakdown of the main components and their functionalities:

1. Speech Recognition Class

• Initialization: The SpeechRecognizer class is initialized with a transcriber instance, sample rate, buffer size, and silence threshold.
• Audio Input: The system captures audio input through the sounddevice library. It continuously listens for audio and stores it in a buffer.
• Voice Activity Detection: Using Silero VAD, the system detects whether speech is present in the captured audio. If speech is detected, it records the audio segment.
• Managing Silence: When silence is detected, the system saves the recorded audio to a .wav file and starts a transcription thread.
• File Management: The recorded audio segments are saved in the recordings/ directory, and each segment is named incrementally.

2. Transcriber Class

• Initialization: The Transcriber class is initialized with a model name, sample rate, and directory for transcripts.
• Transcribing Audio: The transcriber takes each recorded audio file and processes it in chunks using the Faster-Whisper model. Each chunk is transcribed to text.
• Saving Transcripts: Transcriptions are saved in the transcripts/ directory. If a transcript file already exists, a new file is generated with an incremented filename.
• Audio Processing: The audio is loaded from the saved .wav files, and the transcription is performed in manageable chunks for efficiency.

3. Combining and Cleaning Up

• Exit Handler: When the user exits the program (using Ctrl+C), the handle_exit function is triggered. This function combines all individual audio files into a single output file and deletes the original files.
• Directory Structure: The program creates necessary directories if they do not exist, ensuring organized storage of recordings and transcripts.

Usage Flow

1. Start Recording: The user starts the program, which initializes the transcriber and speech recognizer.
2. Capture Speech: The system listens for audio, detects speech, and records it in real-time.
3. Transcription Process: As each audio segment is recorded, a transcription thread is initiated, and the transcribed text is saved to the appropriate file.
4. Exit & Cleanup: On exiting, the recorded audio segments are combined, and the original files are deleted to free up space.

This system efficiently combines real-time audio capture, intelligent speech detection, and automated transcription, making it a powerful tool for various applications such as note-taking, meeting transcription, and more.

Credits

• Silero VAD by snakers4/silero-vad - For efficient voice activity detection.
• Faster-Whisper by [SYSTRAN(https://github.com/SYSTRAN/faster-whisper) - Enables real-time transcription.
• pydub - For handling and processing audio files.

License

This project is licensed under the Creative Commons NonCommercial (CC BY-NC)** License.. Feel free to modify any part of it according to your needs! Let me know if there's anything else I can assist you with. If you want to contribute feel free.