Whisper-tiny-quiztest
| Property | Value |
|---|---|
| Parameter Count | 37.8M |
| License | Apache 2.0 |
| Base Model | openai/whisper-tiny |
| Tensor Type | F32 |
| Best WER Score | 55.05 |
What is whisper-tiny-quiztest?
Whisper-tiny-quiztest is a specialized automatic speech recognition (ASR) model fine-tuned from OpenAI's Whisper-tiny base model. It's specifically optimized for quiz-related speech recognition tasks, trained using the tutikentuti/quiztest dataset.
Implementation Details
The model was trained using PyTorch and Transformers framework, implementing a cosine learning rate schedule with restarts. Training utilized an Adam optimizer with carefully tuned hyperparameters (β1=0.9, β2=0.999, ε=1e-08) and a learning rate of 3e-05.
- Training conducted over 1000 steps with 1000 warmup steps
- Batch size of 8 for both training and evaluation
- Achieved final validation loss of 0.0947
- Uses Safetensors for model storage
Core Capabilities
- Automatic Speech Recognition optimized for quiz content
- Achieved 55.05 WER (Word Error Rate) on evaluation
- Supports real-time transcription through inference endpoints
- Compatible with TensorBoard for monitoring
Frequently Asked Questions
Q: What makes this model unique?
This model is specifically fine-tuned for quiz-related speech recognition, making it particularly suited for educational and assessment applications. Its relatively small size (37.8M parameters) makes it efficient while maintaining reasonable accuracy.
Q: What are the recommended use cases?
The model is best suited for transcribing quiz-related audio content, educational materials, and assessment scenarios where speech recognition is needed. It's particularly useful in applications where computational resources are limited due to its compact size.
