PASSENGER IN-CAR COMMUNICATION ENHANCEMENT Klaus Linhard and Juergen Freudenberger DaimlerChrysler AG, Research & Technology, Dialogue Systems Wilhelm-Runge-Str. 11, 89081 Ulm, Germany phone: +49 731 505 2170, fax: +49 1731 505 4105, email: klaus.linhard@daimlerchrysler.com ABSTRACT The expression passenger in-car communication refers to the conversation (speech communication) of the car passengers while driving. In particular, the communication between dif- ferent seat rows is difficult due to road noise and the acoustic situation within a car. A system which enhances the speech intelligibility is desirable for safety and comfort reasons. We discuss the basic building blocks and design issues of such an in-car communication system, particularly with regard to the acoustic conditions in a car. 1. INTRODUCTION Although luxury vehicles provide a noise reduced environ- ment in the car the communication between the front seats and the rear seats is difficult. In contrast to a normal con- versation, in cars we have road noise and the positions of the passenger seats are fixed which impairs speech understand- ing. Usually passengers do not feel comfortable to conduct long conversations. Frequently, the car driver is tempted to turn the head in order to improve the communication. Thus for safety and comfort reasons, a system which supports nat- ural communication between passengers is desirable. In this paper we discuss an in-car communication system which im- proves speech communication in a vehicle. Such a system basically works as an intercom between the different passen- ger seats. Furthermore, it can serve as the acoustic front-end for other applications like hands-free telephony, voice con- trolled devices, broadcast services, and dialog systems. Sim- ilar concepts are considered for example in [1], [2], and [3]. The system under consideration consists of dedicated micro- phones and loudspeakers (one unit for each passenger), and a digital signal processing part. We discuss the basic design issues for such a system. For example, an in-car communica- tion system operates in a closed electro-acoustic loop similar to public address systems which causes stability problems. In order to provide a full-duplex communication echo can- cellation is required. Moreover, due to the short delays for wave propagation, the tolerable processing delays are also very small. These conditions are in fact similar to the situ- ation with digital hearing aids [4]. In section 2, we discuss the influence of the microphone/loudspeaker positioning on the achievable gains in terms of signal to noise ratios. In the subsequent section, we describe the basic signal processing components required for in-car communication. That is, we consider the necessary efforts in order to stabilize the system as well as for echo cancellation. The discussions are sup- ported by acoustic measurement results obtained in a Mer- cedes S-Class. Some speech samples will be demonstrated at the conference. 95 cm rear mic. 80 cm 120 cm front mic. Fig. 1: Geometric situation for four passengers inside a Mercedes S-Class. 2. MICROPHONE/LOUDSPEAKER POSITIONING In this section, we present measurement results for speech signal propagation within a car. In particular, we consider the transfer characteristics of speech signal propagation between talker and listener depending on the speaker position as well as on the positioning of the microphones and loudspeakers. The corresponding results are helpful for estimating the po- tential gains of in-car communication systems. 2.1 Transfer functions depending on the microphone po- sition Fig. 1 is an illustration of the geometric situation for four passengers inside a car. We give the approximate distances to each other in cm and depict two possible positions for mi- crophones in the front and in the rear of the car cabin. The best solution to capture the speech source would be with a close talk microphone for each passenger (not shown in Fig. 1). Such a close talk microphone is not suitable for practical applications. However, we will use close talk sig- nals as a reference in order to discuss potential gains for in- car communication. That is, in the following measurements we refer to the signal spectrum of a close talk microphone such that the corresponding spectrum is white with a signal level of 0 dB. In Fig. 2 we show measurement results for two speaking situations: driver is speaking and right hand side passenger in the rear is speaking. Because levels at left and right ears are different we give the maximum, respectively. These measures were conducted with frequency linear omni- directional microphones and the frequency averaging is 1/3 octave. We observe from Fig. 2 that the received levels at the 21