ELEC-E5510 - Speech Recognition D, Lecture, 25.10.2023-8.12.2023

Topic outline

• Lectures
  In this list, the 2022 slides will be replaced by the 2023 ones after each lecture is given at latest. The titles may be identical, but the contents are improved each year based on feedback. The project works and their schedule changes each year. 

  For practicalities, e.g. regarding to the Lecture Quizzes and Exercises, check MyCourses > Course Practicalities
  

  • Lectures 1-2 Feature extraction and modeling
    

  • 

    Lectures 1-2 slides (2023) File

    PDF document

    • course organization
    • what is ASR
    • features of speech
    • MFCC
    • GMM
    • DNN
      
  • 

    Lecture 1-2 exercise: Gaussian mixture model Assignment

    Not available unless any of:

    • You are a(n) Student
    • You are a(n) Teacher
    • You are a(n) Teacher (MC)

    Instructions can be found in the pdf file. Please upload your answer here, e.g. as a photo, text or pdf file
    

  • Lectures 3-4 - Phoneme modeling

  • 

    Lecture 3-4 slides (2023) File

    PDF document

    • Phonemes
    • HMMs
  • 

    Lecture 3: exercise Forward Assignment

    Not available unless any of:

    • You are a(n) Student
    • You are a(n) Teacher (MC)
    • You are a(n) Teacher

    Please type or upload your calculations here, e.g. as a photo, text or pdf file to earn a lecture activity point.
    

  • 

    Lecture 3: exercise Viterbi Assignment

    Not available unless any of:

    • You are a(n) Student
    • You are a(n) Teacher (MC)
    • You are a(n) Teacher

    Please type or upload your calculations here, e.g. as a photo, text or pdf file to earn a lecture activity point.
    

  • Lectures 5-6 - Language Modeling

  • 

    Lecture 5 slides (2023): N-gram language models File

    PDF document

    • lexicon
    • language modeling
    • n-grams, smoothing
  • 

    Lecture 6 slides (2023): Neural network language models File

    PDF document

    • Intro to NNLM
    • Recurrent neural network language models
    • Long Short-Term Memory language models
    • Attention

    
    

  • 

    Lecture 6 NNLM exercise Assignment

    Not available unless any of:

    • You are a(n) Teacher (MC)
    • You are a(n) Teacher
    • You are a(n) Student

    Please type or upload your calculations here, e.g. as a photo, text or pdf file to earn a lecture activity point.
    

  • Lectures 7-8 - Continuous speech and decoding
    

  • 

    Lecture 7-8 slides (2023) File

    PDF document

    • recognition in continuous speech
    • token passing decoder
    • improving the recognition performance and speed
      
    • measuring the recognition performance
      
  • 

    Lecture 7 exercise: Token passing decoder Assignment

    Not available unless any of:

    • You are a(n) Teacher (MC)
    • You are a(n) Student
    • You are a(n) Teacher

    Fill in the last column with final probabilities of the tokens, select the best token and output the corresponding state sequence!
    

    The goal is to verify that you have the learned the idea of the Token passing decoder. The extremely simplified HMM system is almost the same as in the 2B Viterbi algorithm exercise. The observed "sounds" are just quantified to either "A" or "B" with given probabilities in states S0 and S1.  Now the task is to find the most likely state sequence that can produce the sequence of sounds A, A, B using a simple language model (LM). The toy LM used here is a look-up table that tells probabilities for different state sequences, (0,1), (0,0,1) etc., up to 3-grams.
    

    Hint: You can either upload an edited source document, a pdf file, a photo of your notes or a text file with numbers. Whatever is easiest for you. To get the activity point the answer does not have to be correct.

    
    

  • Lecture 9-10 - End-to-end ASR with deep neural networks
    

    
    
  • 

    Lecture 9 slides (2023) File

    PDF document

    Not available unless: You belong to any group
  • 

    Lecture 10 slides (2023) File

    PDF document

    Not available unless: You belong to any group
  • 

    Lecture 9-10 slides (2022) File

    PDF document

    Not available unless: You belong to any group

    This is 2022, but because the content was quite different (focusing on attention-based encoder-decoder architectures) this maybe worth studying, too.

  • 

    Lecture 9 exercise Assignment

    Not available unless any of:

    • You are a(n) Teacher
    • You are a(n) Student
    • You are a(n) Teacher (MC)
  • 

    Presentations for the weeks 6-7 File

    PDF document

    Here's the presentation schedule as finalized at the last lecture.