5 File Formats

This chapter describes the file types, as well as the file formats used in HARK . The earlier versions of HARK used a variety of file formats, which made it difficult to grasp overall, especially in the analysis of complex binary formats of transfer function files. Because of that, from HARK 2.1, file formats has been simplified by implementing the 2 design policies below.

Lessen the use of HARK special formats and use standardized format if possible.
Utilize libraries that offers file input/output API.

To comply with these policies, standardized format and compounds of standardized format are now being implemented besides Matrix binary, which is the only HARK-specific file format. In addition, a library that supports file input/output functions called "libharkio3" was created for easy file manipulation.

HARK mainly uses the following three formats:

XML: Files that render the position. The extension is .xml
Matrix binary: Files that represent a matrix. The extensions is .mat
Zip: An organization of files mentioned above used for complex file formats such as transfer functions, etc. The extension is .zip

Other file formats used in earlier versions are either integrated to the file formats above or converted into a standardized format.

Table 5.1 shows the list of files that can be configured in the node’s input/output or property settings.

Table 5.1: HARK nodes that uses file I/O

Node name	Where to use	File type	New format	Old format
SaveRawPCM	Output	Raw Audio file	Raw Audio	No change
SaveWavePCM	Output	Wave file	PCM Wave	No change
LocalizeMUSIC	Property	TF(*) for localization	Zip	HGTF binary
SaveSourceLocation	Output	Localization result	XML	Localization result text
LoadSourceLocation	Property	Localization result	XML	Localization result text
GHDSS	Property	TF(*) for separation	Zip	HGTF binary
	Property	Microphone positions	XML	HARK text
	Property	Stationary noise positions	XML	HARK text
	Property	Initial separation matrix	Zip	HGTF binary
	Output	Separation matrix	Zip	HGTF binary
SaveFeatures	Output	Features	Matrix binary	`float` binary
SaveHTKFeatures	Output	Features	HTK format	No change
DataLogger	Output	`Map` data	XML	`Map` text
CMSave	Property	Correlation matrix	Zip	Correlation matrix text
CMLoad	Output	Correlation matrix	Zip	Correlation matrix text
JuliusMFT	Commandline	Configuration	jconf (text)	No change
	in jconf file	AM(*), phoneme list	julius format	No change
	in jconf file	LM(*), dictionary	julius format	No change
harktool	harktool	Sound source positions list	XML	srcinf text
	harktool	Impulse response	PCM Wave	`float` binary

TF: Transfer Function, AM: Acoustic Model, LM: Language Model

The rest of the chapter will discuss the 3 file formats in detail. Regarding Julius format, see the Julius document for reference. Regarding the difference between the file formats from the original Julius , see the discussion on JuliusMFT . For Raw Audio Format and PCM Wave Format, see the discussion on standard format.