Old and New Work in Discriminative Training of Acoustic Models

Explore the evolution of discriminative training techniques for acoustic models in this 37-minute lecture by Daniel Povey from Johns Hopkins University's Center for Language & Speech Processing. Examine both established and cutting-edge approaches to training acoustic models that optimize discrimination between different speech sounds and phonemes. Learn about the theoretical foundations underlying discriminative training methods and discover how these techniques have advanced over time to improve speech recognition accuracy. Gain insights into the practical implementation challenges and solutions in discriminative training, while understanding the impact of these methods on modern automatic speech recognition systems.