Uzbek Speech-to-Text (STT) Project - Navaistt

This project provides Speech-to-Text (STT) functionality for Uzbek language audio files. It utilizes the islomov/navaistt_v1_medium model from Hugging Face.

Description

The core STT model (islomov/navaistt_v1_medium) is optimized for processing audio segments up to 30 seconds long. This project extends its capability to transcribe longer audio files by:

1. Splitting the input audio into 30-second chunks.
2. Processing each chunk individually using the STT model.
3. Combining the transcribed text from all chunks to produce the final transcription.

Model Used

The STT functionality is powered by the islomov/navaistt_v1_medium model available on Hugging Face.

Features

• Transcription of Uzbek language audio.
• Handles audio files longer than 30 seconds through automatic chunking and result aggregation.

Requirements

• Python 3.x
• Libraries specified in requirements.txt (if available). Key libraries likely include:
  • transformers
  • torch
  • torchaudio

Usage

1. Clone the repository (if applicable) or download the main.py file.
2. Install dependencies:

   pip install -r requirements.txt 
   # Or install libraries manually, e.g., pip install transformers torch pydub

3. Code Usage

   if __name__ == "__main__":
       starting_time = time.time()
       audio_file = "audio.wav"
   
       transcriber = NavaiSTT()
       transcription = transcriber.transcribe(audio_file)
   
       print(f"Transcription: {transcription}")
       print(f"Time taken: {time.time() - starting_time:.2f} seconds")

4. Run the script:

   python main.py

For more information, go to the official website.