When Voice Assistants Miss the Mark: The Need for Inclusive Design in Voice Tech

A colleague in the UK recently drew my attention to a TV advertisement run by the automotive company Hyundai that seeks to educate users about how to pronounce its brand name properly. In the ad, different Hyundai customers attempt to use voice technology on their mobile phones to navigate to the nearest Hyundai dealership. To their confusion, their phones direct them to unrelated places like a hair salon called High ‘N’ Dye and an optician called Highland Eye. It isn’t until a bystander corrects the mispronunciation that users realize where they had been going wrong.

This ad is interesting for two reasons. First, it highlights just how critical voice technology is becoming for businesses. Voice technology has completely transformed how we interact with our devices and the digital world. As of 2023, it is estimated that there are over 4.2 billion digital voice assistants in use worldwide, and this number is projected to reach 8.4 billion by 2024 (Statista). As a result of this surge, brands are increasingly reliant on accurate voice recognition to ensure discoverability and engagement.

Second, it is neither users’ nor Hyundai’s fault that its dealerships are difficult to discover through voice search. As is to be expected, users often struggle with uncommon or foreign brand names like Hyundai. Even here in Boston, MA, I suddenly found myself asking colleagues how they pronounce Hyundai and their responses varied. Beyond mispronunciations, regional accents, dialects, and even individual speech patterns can all contribute to inaccuracies in voice recognition. It is the responsibility of the companies developing voice technology to account for these variances.

The High Stakes of Voice Search Accuracy

Voice search is becoming even more entrenched in user journeys. The 2022 Voicebot.ai U.S. Smart Speaker Consumer Adoption Report noted a significant increase in the variety and frequency of commands used with voice assistants, suggesting that users are becoming more reliant on voice technology for a wider range of tasks. At the same time, the report also highlights that users frequently experience frustration with voice assistants, particularly when it comes to understanding complex commands or accents.

This is corroborated by Applause’s own research. The 2023 Voice and AI Survey found that 30% of users are either somewhat or extremely dissatisfied with voice assistants. When asked about their general sentiment towards the technology, the most common answer was “I would use voice assistants more if they responded more accurately to the way I phrase things.” Better understanding of language variations and responses ranked among the top three suggestions for how voice assistants could be improved, alongside more relevant responses and responses more specific to the user.

Good and Bad Strategies for Accuracy and Inclusivity

Right now, I’ve seen examples of companies offering workarounds to mispronunciations that don’t solve the root issue. While providing alternative search options like text input or visual menus offers a fallback for users struggling with voice commands, there’s no guarantee users won’t just give up after voice search fails the first time. The same goes for pronunciation guides; users aren’t realistically going to read these in hectic scenarios and they are just another potential source of frustration.

There is simply no shortcut to fighting mispronunciations in voice technology other than to tackle the issue in the model directly. Brands can’t change their names, but developers can change how well their models take into account variations in pronunciation.

Voice Recognition Solutions for Brands

Voice recognition technology can be enhanced through various approaches:

• Diverse Training Data: By training voice models on a wider range of accents, dialects, and pronunciation variations, we can teach them to better understand the diversity of human speech.
• User Feedback Loops: Continuous feedback from users can help identify and rectify recurring recognition errors, improving the system over time.
• Thorough Testing: Rigorous testing with diverse user groups can reveal potential pronunciation issues before they impact the user experience.
• Regular Updates: Voice models should be updated regularly to incorporate new data and improve accuracy.

The Future of Voice Search Means Embracing Inclusivity

Today, accurately recognizing pronunciations is just the starting point for voice experiences. Research indicates that AI now can classify emotional states with an accuracy rate of up to 85% by analyzing vocal patterns, allowing voice assistants to respond in ways that feel more understanding and kind. Additionally, innovations like multilingual support, voice biometrics for enhanced security, multimodal interactions and personalized responses are shaping a future where voice technology is not only accurate but also inclusive and user-centric.

Voice technology should work for everyone, regardless of their accent, dialect, or pronunciation. The Hyundai advert, while amusing, serves as a reminder that there’s still work to be done.

Video