ASR Evaluation Framework: Benchmarking Speech Recognition Models Across Accuracy, Speed, and Robustness

Picking an ASR model for production is not straightforward. Whisper might be the most accurate for general English but too slow for real-time use. Wav2Vec2 might be fast enough for edge devices but st