User Manual

Getting Started

Introduction to SPRAAK concepts, conventions and file structures

• Features of the SPRAAK Toolkit
• Linguistic Definitions, Conventions and Files
• Databases & Segmentations
• Feature Extraction
• Acoustic Model
• Language Model
• Using standalone SPRAAK routines

Tutorials and Reference Experiments on TIMIT

• Introduction
• Recognizing and Aligning with SPRAAK
• Baseline Acoustic Model Training
• Vocal Tract Length Normalization (VTLN)
• MIDA & Gaussian Decorrelation
• Context-Dependent Model Training

Reference Experiments for WSJ

• Introduction
• Preparing the data
• HMM Based Training
• HMM Evaluation and Fine Tuning
• Example Based Recognition on WSJ database

Extension to a Missing Data Theory based recognizer

• Background
• Data components for the acoustic model
• Mask estimation
• Example on the Aurora 4 database

Exploring Further

• Building ASR Applications with SPRAAK

Frequently Asked Questions