Deep Learning for Speech and Language Processing

Course Overview

The Deep Learning for Speech and Language Processing course provides a comprehensive introduction to modern deep learning techniques applied to speech recognition, text processing, and natural language understanding. Students gain practical experience with Python, using both PyTorch and TensorFlow to implement and train deep learning models.

Topics Covered

Throughout the semester, we explored key topics in deep learning for speech and language, including:

• Fundamentals of Deep Learning: Neural networks, optimization, and regularization techniques.
• Word Embeddings & Representation Learning: Word2Vec, GloVe, FastText, and contextual embeddings like BERT.
• Recurrent Neural Networks (RNNs) & Attention: LSTMs, GRUs, self-attention mechanisms, and the Transformer architecture.
• Speech Processing: Feature extraction (MFCCs, spectrograms), automatic speech recognition (ASR), and end-to-end speech models.
• Natural Language Processing (NLP): Sentiment analysis, named entity recognition (NER), sequence labeling, and text classification.
• Sequence-to-Sequence Models: Machine translation, speech-to-text, and text-to-speech (TTS).
• Pre-trained Language Models & Transfer Learning: BERT, GPT, and fine-tuning strategies.

Programming & Assignments

Students implemented deep learning models using PyTorch and TensorFlow, applying them to real-world speech and language tasks. Assignments included:

• Building and training word embedding models.
• Implementing RNNs, LSTMs, and Transformers for text classification.
• Developing an ASR pipeline for speech recognition.
• Exploring pre-trained language models for NLP tasks.

Course Materials

Lecture slides, assignments, and additional resources are available on the course webpage.