My notes for Prof. Hung-Yi Lee's lectures on Deep Learning for Human Language Processing. I really enjoy the lectures, I hope my notes are useful.
This course teaches about using deep learning to let computer gain the ability to interpret human language, writes or speaks in a way that human is capable of understanding
Why is it call Human Language Processing (HLP) instead of Natural Language Processing (NLP) ?
- Human language exists as speech or text
- Only 56 % languages have written form
- Most NLP courses and books focus mostly on texts
- In this course, the ratio of focus on text and speech is 5:5
The following table is a oversimplified summary of the coverage of this course :
Input | Output | Application |
---|---|---|
Audio | Text | Speech recognition |
Audio | Audio | Speech separation, voice conversion |
Audio | Class | Speaker recognition, keyword spotting |
Text | Audio | Text-to-Speech synthesis |
Text | Text | Chat-bot, translation, summarization, question answering |
Text | Class | Sentiment analysis |
- Speech Recognition
- Voice Conversion
- Speech Seperation
- Speech Synthesis work in progress
- Speaker Verification work in progress
- Natural Language Processing Tasks
- BERT and its family work in progress
- Non-Autoregressive Generation work in progress
- Text Style Transfer work in progress
Thanks to Prof. Hung-Yi Lee and all the Teaching Assistants of this course for providing knowledge for free
These notes are based on my understanding of Prof Lee's lectures. It might consists of mistakes. For clarity, please go through the presentation slides.