Case study for Top Automotive OEM

The Challenge

A leading global automotive electronics software provider needed our team to help provide them with the audio language data needed in their in-vehicle speech recognition system.

To complete the build of this system, the ability to recognize and correctly process voice commands is an essential speech data training capability. This speech data changes as the way people speak changes. Drivers will also use a variety of different spoken expressions to adjust various aspects of the car such as temperature control, broadcast volume, navigation commands, and phone calls. The multilingual and multi-dialectal nature of the training process, as well as the multiple spoken language standards, posed the biggest challenge for this task, which required us to provide thousands of expressions for the training data for all types of content.

The Solution

Our team's extensive resources helped us quickly recruit the native speakers needed for the project to capture the various recordings for the given scenarios. We had a professional tts team to control the quality of the recordings to strict standards, and professional linguists to guide the quality standards of the language to better match the language specifications of the automotive industry.

In the process of collecting voice data, we only give the content of the specified topic without any pre-determined scripted information, for example, how to lower the temperature and other specific needs in a given set of scenes, and then give various corresponding. It is completely in the state of unscripted, spontaneous speech collection.

In addition, our text data collection is also equipped with professional scripts for voice data collection of fixed words. By realistically simulating the state of the driving environment, the person being collected can respond more naturally and realistically.

Results

With our professional team providing guidance and training, we were able to collect speech data that met the client's needs, the language diversity was fully guaranteed, and the company quickly expanded the development of more than 40 language recognition systems with our help. The high quality and large amount of training data effectively improved the efficiency capability of all stages of model development.