The ANIR Corpus

   
  Contents:  
   
   
   
   
   
   
   
 

The "ANIR" (Automotive Noise and Impulse Responses) corpus allows simulating realistic acoustic environments inside automotive vehicles. With the data contained in this corpus it is possible to test and develop systems for speech-based services, such as hands-free und in-car communication (ICC), without having an equipped car available. For this purpose, background noise is recorded in different driving situations. Furthermore, a large number of impulse responses are measured for combinations of louspeakers and microphones inside the vehicle. Specifically, loudpeakers can either be the mouth loudspeakers of head and torso simulators (HATS) or regular car loudspeakers. Correspondingly, microphones are either the HATS ear microphones or the hands-free/ICC microphones.

Several HATS are placed on different seats in the car (as you can see in the picture on the left) to conduct the measurements. For vehicles with a large number of seats, puppets with binaural headsets for aurally accurate sound recordings are used in addition. Such a headset is also used by the driver of the vehicle during the recording of the background noise. All measurement channels are properly calibrated and synchronized.

So far, measurements have been conducted in a Daimler V-Class. In the future, the corpus will be extended with measurements from other vehicles.

 

Key Features

Several background noises are recorded in different driving situations. Here, mainly the speed of the car and the level of the air condition varies. For the measurements several microphones and loudspeakers are used. For this, microphones/loudpspeakers added by us in a custom retrofit may be used in addition to the ones built in by the manufacturer. Of course, head and torso simulators with mouth loudspeakers and ear microphones are used as well. Next to the noise recordings, the impulse responses from every loudspeaker to every microphone are measured. Currently, this has only been done in a Daimler V Class, but in the future the recordings and measurements are to be replicated in different cars.

 

State of the Corpus

On this page you will find an overview showing the cars in which measurements have been conducted. There, you can also find the different driving situations (conditions) in which background noise was recorded. Currently, the corpus contains over 600 impulse responses and over 500 background noise recordings of the Daimler V class.

 

Download

Here you can download the background noise recordings and the impulse responses as a zip folder. We have organized the files in different groups, so you can choose whether you want to download all files or only those of a specific car. You can also separately download the impulse responses or the noise files of a specific car. Furthermore, we provide example code for Matlab and Python to read the impulse responses.

 

Examples

 

{tab-cars V Class (Daimler)}

Background Noise

 

{tab-background-noise Standard Car Microphone - 0 km/h}

Mic. 0 - Speed 0 km/h (engine on), air condition at level 1 (low level)

Play (normalized) sound  

{tab-background-noise Ear Microphone - 50 km/h}

Mic. 21 - Speed 50 km/h (engine on), air condition at level 7 (high level)

Play (normalized) sound  

{tab-background-noise Standard Car Microphone - 100 km/h}

Mic. 10 - Speed 100 km/h (engine on), air condition at level 4 (middle level)

Play (normalized) sound  

{/tabs-background-noise}

 

Impulse Responses

 

{tab-impulse-responses Headrest Lsp. 0 to Standard Mic. 0}

Left headrest Loudspeaker of Zone 1 (Lsp. 0) to Standard Car Microphone 0 (Mic. 0)

{tab-impulse-responses Mouth Lsp. 14 to Ear Mic. 23}

Mouth Loudspeaker of Zone 3 (Lsp. 14) to Left Ear Microphone of Zone 3 (Mic. 23)

{tab-impulse-responses Standard Lsp. 21 to Standard Mic. 18}

Standard Car Loudspeaker 3 (Lsp. 21) to Standard Car Microphone 18 (Mic. 18)

{/tabs-impulse-responses}

{/tabs-cars}

 

Creators of the Data Base

The data base was created as a joint work between Kiel University (CAU). Involved researchers are:

 

Corresponding Publications

     

T. Hübschen, M. Gimm, G. Schmidt: A Background Noise and Impulse Response Corpus for Research in Automotive Speech and Audio Processing, Proc. DAGA, Germany, 2022