1 Introduction

This document describes frequently occurring problems and their solutions. We call each pair, consisting of a problem and its solution, a “recipe”, analogous to a recipe for cooking.

Chapter 2: Learning HARK

This chapter includes recipes for those beginning HARK. These recipes describe methods of recording sounds, and localizing, separating, and recognizing. If this is the first time you are using HARK, we recommend that you read the recipes in this chapter.

Chapter 3: Something is Wrong!

This chapter describes troubleshooting common problems including installation and recording. It also includes debugging recipes. If you have trouble, check this chapter.

Chapter 4: Microphone Array

This chapter includes recipes for designing microphone arrays, including, for example, the number and kinds of microphone to use and how to install a microphone in a robot.

Chapter 5: Input Data Generation

This chapter includes recipes for generating input data for HARK by recording and simulation. These include, for example, multichannel recording, impulse response measurements, and multichannel sound generation by simulation.

Chapter 6: Acoustic and Language Models

The speech recognizer Julius , which is supported by HARK, requires an acoustic model and a language model. This chapter describes how to build these models.

Chapter 7: FlowDesigner

In HARK, you build a robot audition system using FlowDesigner middleware. FlowDesigner includes the placement and connection of nodes. This chapter includes tips for using FlowDesigner . See another document for HARK Designer, a newly developed GUI for building a network from Ver 1.9.9.

Chapter 8: Sound Source Localization

This chapter includes recipes for sound source localization including localization system building, debugging, and parameter tuning.

Chapter 9: Sound Source Separation

This chapter includes recipes for sound source separation including separation system building, debugging, and parameter tuning.

Chapter 10: Feature Extraction

Speech recognition requires the extraction of features from separated sound. This chapter includes recipes for the introduction to features used for speech recognition and how to extract them. This chapter also includes the recipes of Missing Feature Theory used to select reliable features.

Chapter 11: Speech Recognition

This chapter includes how to make a configuration file for Julius .

Chapter 12: Others

This chapter includes miscellaneous recipes, such as selecting a window function for frequency analysis.

Chapter 13: Advanced recipes

This chapter includes advanced recipes, such as those for adding a new function to HARK and connecting HARK to other systems.

Chapter 14: Sample Networks

This chapter includes sample network files.