Efficient Speech Processing

Explore cutting-edge research in speech processing through this CLSP Seminar talk by Tatiana Likhomanenko. Dive into three key areas of speech technology: recognition, generation, and privacy. Learn about self-training algorithms for semi-supervised learning in large-scale speech recognition and discover strategies to make them simple and resource-efficient. Examine how error correction models can outperform conventional language models at scale. Investigate the joint modeling of text-to-speech and speech-to-text using a simple tokenization scheme. Gain insights into training speech recognition systems with differential privacy and federated learning. Benefit from Likhomanenko's extensive experience in machine learning, including her work at Apple, Facebook AI Research, and her background in physics and mathematics.