A training set selection competition for keyword spotting
Big Data is expensive. Training on hundreds of thousands of recordings takes time, money and energy. What if you could train a model and achieve similar performance with a fraction of the data? If this sounds interesting, stay here!