Kinyarwanda-speech-to-text-ASR/Kinyarwanda-speech-to-text-ASR
Domain:
natural language processingRecord type:
model
Creator:
Kin
Host:
Kinyarwanda Automatic Speech Recognition Model based on Whisper
# KinyaWhisper - Kinyarwanda Automatic Speech Recognition Model
## 🌟 Project Overview
KinyaWhisper is an automatic speech recognition (ASR) model specifically optimized for Kinyarwanda, fine-tuned based on OpenAI's Whisper architecture. This project is dedicated to providing high-quality speech recognition solutions for low-resource languages, particularly offering advanced AI speech technology support for the Kinyarwanda-speaking community.
## 🎯 Model Capabilities
### Core Features
- **Automatic Speech Recognition**: Convert Kinyarwanda audio to text
- **Multi-scenario Adaptation**: Support speech recognition for different age groups and genders
- **Real-time Processing**: Support batch and individual audio file processing
- **High-precision Recognition**: Professionally trained and optimized for Kinyarwanda
### Technical Specifications
- **Base Model**: OpenAI Whisper-Small
- **Language Support**: Kinyarwanda
- **Audio Format**: WAV (16kHz sampling rate)
- **Maximum Audio Length**: 30 seconds
- **Feature Extraction**: 80-dimensional Mel spectrogram
- **Model Size**: Approximately 244MB
### Performance Metrics
- **Best Word Error Rate (WER)**: 20.07%
- **Training Dataset**: 90,162 training samples
- **Validation Set**: 4,649 samples
- **Test Set**: 4,634 samples
- **Test Set Accuracy**: 85%
- **Validation Accuracy**: Best performance achieved at epoch 4
- **Processing Speed**: Supports GPU-accelerated inference
## 👥 Development Team
### Project Leader
**Lin Fan** - Lead Developer & Project Manager
- **Institution**: University of Bristol, UK
- **Education**: Master's in Computer Science (in progress)
- **Department**: School of Engineering, Computer Science
- **Contact**: 2978831945a@gmail.com
- **Responsibilities**: Overall project architecture design, technical solution development, team coordination
### Core Technical Team
#### Data Engineering Team
Responsible for Kinyarwanda speech data collection, cleaning, preprocessing, and feature engineering to …
Similar
Kinyarwanda Agricultural Text-to-Speech DatasetAV55CS/Speech-to-Text-Fine-tuning-Whisper-Tiny-for-Swahili-ASRIshita95-AIML/Kinyarwanda-Automatic-Speech-Recognition-Track-A-An-ASR-zakariakassemi2000/Application-Flutter-Darija-Speech-to-Text-Text-to-Speechspeech-recognition123/Speech-to-Textnextgencodex1/kinyarwanda-ASR