14.1.1 Category of sample network

All categories of sample networks are shown in Table 14.1. There are five categories. The sample networks of each category and files required for execution of the samples are stored in the directories indicated the sample directories on the right column in the table.

For execution of a sample network, execute a script corresponding to each network file name. Set values of arguments to be given to the network file and necessary setting files are described. For example, if there a network file named demo.n, the script file name that executes the network is demo.sh. Character strings before "." are commonized. However, demo.sh may include the setting items that depend on operating environments and therefore unexpected result may occur when executing without confirmation. The correct setting method is described in description of each sample network.

Indicate below outline of category of each sample.

• Sound recording network
  This is a sample for which modules of the AudioIO category of HARK are used. Monaural sound recording and stereophonic recording are included as basic sound recording samples Stereophonic recording and monaural sound recording operate in most hardware environments.

• Sound source localization network
  This is a sample for which modules of the Localization category of HARK are used. This is a sample in particular for usage of LocalizeMUSIC . A sample in which a sound source localization result is displayed on a screen with DisplayLocalization and saved in a file with SaveSourceLocation is available. Recorded sounds for eight channels is available so that sound source localization processing can be confirmed off-line/ Since it is off-line processing, AD/DA is not required and therefore any computers can operate if HARK is already installed. In order to execute online localization processing, AD/DA for the multi-channel recording that HARK supports is required.

• Sound source separation network
  This is a sample for which modules of the Separation category of HARK are used. This is a sample in particular for usage of GHDSS and PostFilter or GHDSS and HRLE . A sample in which off-line sound source localization processing is performed to recorded sounds for eight channels is available. Since it is off-line processing, AD/DA is not required and therefore any computers can operate if HRAk is already installed. In order to execute online localization processing, AD/DA for the multi-channel recording that HARK supports is required.

• Acoustic feature extract network
  This is a sample in which modules of the FeatureExtraction category of HARK are used. This is a sample in particular for usage of MSLSExtraction and MFCCExtraction . A sample in which off-line acoustic feature extract is performed to recorded sounds for one channel is available.

• Speech recognition network
  This is a sample in which ASR of HARK and modules of the MFM category are used. This is a sample in particular for usage of MFMGeneration and SpeechRecognitionClient . A sample in which off-line sound source localization processing is performed to recorded sounds for eight channels is available.