LISTEN HRTF DATABASE |
CONTEXT | SYSTEM & PROTOCOL | DOWNLOAD | TOOLS |

DEMO SOUNDS | LINKS | GLOSSARY | BIBLIOGRAPHY |

About HRIR,we release one archive per
subject. In each archive, you could find raw and compensated data, in both

Micro$oft(TM) WAV and Matlab(TM) MAT formats.

For each subject, we have collected many informations
gathered in one XML file.

Impulse responses sampling rate is 44100 Hz, measurement length is 8192 points, and quantification is 24 bits.

For each subject, there are 187 stereo 24 bits WAV files, one for each
position of the source. The generic name is : IRC_<subject_ID>_<status>_R<radius>_T<azimuth>_P<elevation>.

Name | Example |
---|---|

<subject_ID> | from 1000 to 1999 |

<status> | R for raw data |

<radius> | radius in centimeters (4 digits) |

<azimuth> | azimuth in degrees (3 digits, from 000 to 180 for source on your left, and from 180 to 359 for source on your right) |

<elevation> | elevation in degrees, modulo 360 (3 digits, from 315 to 345 for source below your head, 0 for source in front of your head, and from 015 to 090 for source above your head) |

These files are stereo HRIR; it is a time domain description.

**Example** : in IRC_1003_R/ directory, the file IRC_1003_R_R0195_T270_P345.WAV
contains the left and right impulse response, in its rough state; source
is at a distance of 1.95 meter, on the right (270 degres) and below your
head (-15 degres).

For each subject, there is only one Matlab file containing a structure. Matlab structure data type seems to be a good choice, in order to gather in one variable both position information and signal data. We use the following conventions :

elev_v | : | a column vector of elevations in degrees |

azim_v | : | a column vector of azimuths |

content_m | : | the data matrix (one line per position) |

type_s | : | a keyword which indicates the content type (FIR) |

sampling_hz | : | sampling rate (in hertz) |

elev_v and azim_v fields contain position information, and are the same
number of lines as the .content_m data matrix. elev_v elements belong to
[-90 ; 90], and must be sorted in increasing order. For each elevation,
the set of azimuths in azim_v vector is sorted in increasing order. Finally,
elev_v and azim_v look like this :

elev_v | azim_v |

-45 | 0 |

-45 | 15 |

... | ... |

-45 | 345 |

-30 | 0 |

-30 | 15 |

... | ... |

-30 | 345 |

... | ... |

... | ... |

75 | 345 |

90 | 0 |

Equalization is performed thanks to diffuse field. Its principle consists in using a reference soundfield to equalize the HRIR. This equalization technique is called "decoupled equalization" [Blauert97,Larcher98], because the recording system and the reproduction system can be equalized in independent sessions using different equipment : the effect of the amplifier, microphones and ears canals are eliminated locally in each session.

Only the magnitude is equalized: the phase is not taken into account. Diffuse field HRTF's magnitude spectrum is approximated by the power average of HRTF measured in free field, i.e. in an anechoic room, for a set of incidences as complete as possible (all the 187 directions, for each ear separately). As the measured directions do not sample the sphere uniformly, the HRTF should be given a weight in the averaging process, proportional to a solid angle associated to the correspondig direction.

The main steps are as follow :

- windowing of the HRIR so that only the same 512 points are kept (the propagation delay is removed)
- moving in the Fourier domain, and split of the transfer function in minimum phase and all pass components
- normalization of the minimum phase by the corresponding diffuse field HRTF
- construction of an equalized transfer function with the equalized minimum phase and the original all pass components
- moving in the time domain

This directory contains same number of files than RAW WAV directory, and the
generic name is the same except the <status> field which is set to C.

For each subject, there is only one Matlab file containing a structure, same skeleton as RAW MAT.