Boosting In-Cabin Speech Recognition: Data Collection

The Challenge:

A leading automotive software provider approached us with the challenge of collecting speech data to enhance their in-cabin speech recognition capabilities. Training the system to accurately recognize and process voice commands required a large and diverse dataset. However, variations in how people speak, including dialects, accents, and tones, presented a significant challenge. Moreover, the complex background noise within the vehicle impacted the accuracy of the recognition system.

The Solution:

We offer comprehensive natural speech data collection services, covering various scenarios and linguistic variations. With the guidance of our language experts, we efficiently collect speech data for specific scenarios in different languages and locations. Using professional equipment, we simulate diverse environmental conditions without scripted dialogue, ensuring authenticity. Native speakers from multiple countries participate in scene simulations, allowing us to capture the drivers' emotional states and other relevant elements, while managing languages relevant to the automotive industry.

The Result:

Through our data collection and annotation services, the company successfully expanded its system to support over 30 languages, improving their in-cabin speech recognition capabilities.