<?xml version="1.0" standalone="yes"?>
<Paper uid="H01-1047">
  <Title>Linguatronic Product-Level Speech System for Mercedes-Benz Cars</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4. TECHNOLOGY
</SectionTitle>
    <Paragraph position="0"> The recognizer used in Linguatronic is speaker independent HMM-based. For the user definable telephone directory, an additional DTW recognizer is used. A key issue for speech technology in the car, noise reduction, is achieved by an integrated signal pre-processing that reduces the influence of noise and other sources of interference to improve the quality of the recognition. Furthermore, the system automatically and imperceptibly adapts itself to the characteristics of the speaker's voice or the acoustic background within vehicles. So, the key technology features are:  Now, from the recognizer side, that doesn't seem too much different to what other speech companies and research groups offer. Yet, recognition accuracy is at a very high level. What is it, then, that makes the difference? Obviously, training of speech data plays a major role here, but others (e.g. the SpeechDat-Car project) have done major data gathering exercises as well. So, we'll point here to some items that are not naturally in the focus of the scientific community, viz. integration, co-operation and engineering.</Paragraph>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
5. INTEGRATION
</SectionTitle>
    <Paragraph position="0"> All upper-level cars of the major European makers these days are equipped with a data bus system, sometimes still over copper wire, but mostly using fiber optics. The Temic system supports the most common automotive bus systems like D2B, CAN, I-Bus and in the near future MOST. This means that a multitude of devices on the bus can be addressed and operated using a single and uniform speech dialogue module, and there is no need to put extra wiring in the car. Furthermore, the push-to-activate (PTA) button, muting of the audio and use of the car's loudspeakers are facilitated, and, very important, the same microphone can be used for both speech commands and telephone. The quality of the microphone (or an array), its characteristics and its positioning are extremely important.</Paragraph>
  </Section>
  <Section position="6" start_page="0" end_page="0" type="metho">
    <SectionTitle>
6. CO-OPERATION
</SectionTitle>
    <Paragraph position="0"> This kind of close integration can only be achieved in an equally close co-operation with the car manufacturer. The speech Command&amp;Control has to fit smoothly into the overall humanmachine-interface for the driver. From the position of the PTA button or lever, via the cancellation function to the analogous behavior of speech and tactile operation of devices, everything has to be done to ensure that speech is an integrated and natural part of the interaction. Fallback modes must be made available in case speech fails. The auto maker must also be aware that even if recognition were one-hundred-percent accurate, people do not always say what they mean or know what they can say, so there will be failures.</Paragraph>
  </Section>
  <Section position="7" start_page="0" end_page="0" type="metho">
    <SectionTitle>
7. ENGINEERING
</SectionTitle>
    <Paragraph position="0"> And, of course, the car manufacturer also knows which price people are willing to pay. The key issue now is the engineering task to deliver a product with a certain quality standard for a certain price. Besides software solutions, Temic currently offers two harware versions of its Command&amp;Control system for cars, one as a plug-in board the size of a credit card, the other as a self-contained box that interfaces directly to the bus and does not require any host processor. Both not only do the recognition, but also store the speech output for the dialogue. And: at a recognition rate that convinced not only Mercedes-Benz, but also others who have a reputation for quality, Linguatronic, under the label 'voice recognition' sells in the US as part of a mobile phone package for a retail price of $480.00; the majority of integrated phones in Mercedes-Benz cars in the US is ordered with this option. Up to the end of the year 2000, Temic has sold more than 170,000 units for automotive applications..</Paragraph>
  </Section>
class="xml-element"></Paper>