1. Abstract
The Telugu Speech-to-Text System is an AI-powered application that converts spoken Telugu language into written Telugu text using Automatic Speech Recognition (ASR) and Natural Language Processing (NLP) techniques. The system captures audio through a microphone or accepts uploaded audio files, processes the speech, and generates corresponding text. It helps users communicate efficiently, create Telugu transcripts, and improve accessibility for people who prefer speaking rather than typing.
2. Objectives
- To convert Telugu speech into written text automatically.
- To support microphone input and audio file uploads.
- To improve accessibility for Telugu-speaking users.
- To reduce manual transcription effort.
- To generate readable Telugu text from spoken audio.
- To support educational, professional, and personal applications.
3. Existing System
Traditional transcription methods require users to listen to audio recordings and type the spoken words manually. Some speech recognition systems provide limited support for regional languages or struggle with accents, background noise, and variations in pronunciation.
Disadvantages
- Manual transcription is time-consuming.
- Accuracy may decrease in noisy environments.
- Regional accents and pronunciation variations can affect recognition.
- Long audio recordings require significant effort to transcribe.
- Some tools have limited support for Telugu speech.
4. Proposed System
The proposed system uses a pretrained speech recognition model or speech-to-text API that supports Telugu. Users can speak through a microphone or upload an audio recording. The system processes the audio, recognizes the spoken Telugu words, and displays the transcription in Telugu script. Additional features can include text editing, copying, downloading, and audio playback.
Key Features
- Telugu speech recognition.
- Microphone-based voice input.
- Audio file upload and transcription.
- Telugu-script text output.
- Editable transcription results.
- Copy and download options.
- Support for different audio lengths, subject to model limits.
5. Implementation Procedure
- Develop a user interface for microphone input and audio file uploads.
- Capture or load the Telugu audio recording.
- Preprocess the audio by converting it to a supported format and sample rate.
- Integrate a Telugu-compatible ASR model or speech recognition API.
- Process the audio and recognize spoken words.
- Convert the recognized speech into Telugu text.
- Display the transcription in the application.
- Allow users to edit, copy, or download the generated text.
- Test the system with different speakers, accents, speaking speeds, and noise conditions.
- Evaluate performance using Word Error Rate (WER) and manual transcription comparisons.
6. Software Requirements
- Programming Language: Python
- Frontend: Streamlit or React.js
- Backend: Flask or FastAPI (optional)
- Speech Recognition: AI4Bharat IndicConformer, Whisper, or another model with verified Telugu support
- Audio Processing: Librosa, SoundFile, or FFmpeg
- Machine Learning Framework: PyTorch or TensorFlow, depending on the model
- Development Environment: VS Code or Jupyter Notebook
- Operating System: Windows, Linux, or macOS
7. Hardware Requirements
- Processor: Intel Core i3 or equivalent and above
- RAM: Minimum 4 GB; 8 GB recommended
- Storage: At least 5 GB of free space
- Microphone: Required for live speech input
- GPU: Optional for API-based recognition; beneficial for running larger models locally
- Internet Connection: Required for cloud-based speech recognition and model downloads
8. Advantages of the Project
- Converts Telugu speech into text quickly.
- Reduces manual typing and transcription effort.
- Supports Telugu-language digital communication.
- Improves accessibility for users who prefer voice input.
- Helps create transcripts of lectures, meetings, and recordings.
- Can be integrated into voice-enabled applications.
- Supports the preservation and digital use of regional-language content.