M13 · Level II — Machine Intelligence
Language & Speech Intelligence
Derive language and speech intelligence from physical signals and symbolic structure, then build robust sequence, speech and multilingual systems.
Topics
Technical concepts
- 01Language as Structured Symbolic Data
- 02Tokenisation as a Computational Interface
- 03n-Gram Language Models
- 04Cross-Entropy, Entropy and Perplexity
- 05Embeddings and Distributional Representation
- 06Syntax, Composition and Long-Range Dependency
- 07Meaning, Context and Pragmatic Information
- 08Speech as an Acoustic Signal
- 09Speech Sampling, Quantisation and Aliasing
- 10Framing and Short-Time Analysis
- 11Fourier Analysis and the Short-Time Fourier Transform
- 12Mel-Scale Features and MFCC-Style Representation
- 13Phonetics, Phonology and Sub-Word Speech Structure
- 14Language Models Inside Speech Recognition
- 15Connectionist Temporal Classification
- 16RNN-T / Transducer-Style Sequence Alignment
- 17Attention-Based Encoder-Decoder Speech Recognition
- 18Speech Decoding and Beam Search
- 19ASR Evaluation and Error Decomposition
- 20Text-to-Speech as Inverse Perception
- 21Prosody, Timing and Expressive Speech
- 22Speaker and Paralinguistic Information
- 23Multilingual Language Intelligence
- 24Multilingual Speech and Code-Switching
- 25Text Normalisation and Evaluation Interfaces
- 26Language Representations as a Knowledge-System Interface
- 27Language and Speech Robustness
- 28Generative Language Behaviour and Hallucination as an Evaluation Problem
- 29Quantum Technology Bridge: Language, Speech and Hybrid Computation
Practical implementation
- 01Laboratory I: Language Model from First Principles
- 02Laboratory II: Speech Signal Processing Pipeline
- 03Laboratory III: CTC Alignment Engine
- 04Laboratory IV: Tiny ASR System
- 05Laboratory V: Compact Text-to-Speech Chain
- 06Laboratory VI: Multilingual Representation Study
- 07Laboratory VII: Speech Robustness Matrix
- 08Laboratory VIII: Speech-Text Alignment
Tools and technologies
Project
End-to-end language, speech or audio-text system
Expected learning outcomes
- 01Can derive tokenisation and probability interfaces.
- 02Can implement core speech signal-processing operations.
- 03Can explain and implement an alignment objective such as CTC on a reduced problem.
- 04Can build and evaluate a small language model and ASR system.
- 05Can analyse decoding and edit-distance error structure.
- 06Can build a constrained speech-synthesis or audio-text alignment experiment.
- 07Can evaluate multilingual, code-switching and noise robustness.
- 08Can reproduce a language/speech research result with documented deviations.
- 09Can defend all major design choices in a technical viva.
Assessment
- 01Daily practice
- 02Laboratories
- 03Research reproduction
- 04Integrated project
- 05Technical viva
- 06Engineering record
Source document: TUTOR Frontier Module 13 Language and Speech Intelligence