In our never ending quest to unravel the mystery that is voice search, we decided to put our favorite voice search assistants to the ultimate test. 9 languages, 12 devices, 1 dedicated lunch hour.
The idea for an epic voice-to-voice showdown developed in August, after learning that Google had added 30 additional languages to its speech recognition and dictation roster. Now supporting 119 language and counting, we thought this was the perfect opportunity for Google to put its money where its mouth is. But would this new dictation support cross over into voice search?
As we were floating ideas around how we could test voice search literacy, Apple had its extravagant keynote event at the brand-spanking-new campus in Cupertino, CA. Between the announcements of the pricey new iPhone X, as well as Apple TV and Watch upgrades, iOS11 was also released – and with it some big upgrades to Siri. Using the power of machine learning, Siri will now answer your questions with a more natural, expressive voice.
Siri can also:
- Understand the context of related queries
- Learn your musical taste and offer recommendations (with Apple Music)
- Translate US English to French, German, Italian, Spanish and Chinese – but not the other way around (this feature is still in beta and expected to improve over time)
The recent updates to the biggest voice search assistants in the game meant this was the perfect time to test out their language literacy with our incredible team and all of the languages they represent. Let the battles begin!
Round 1: Accent Battle
How well do voice assistants process the many different forms a language can take? In our accent battle, a few variations of the English languages were represented: British English, Canadian English, and New Zealand English.
Our questions took into consideration some of the most popular uses for voice search:
Winner: Canadian English
It was a tight race to last place, but in the end Canada came out on top with one correct answer by Siri!
Best (Worst) Answer:
Round 2: Dialect Battle
Similar to the accents, different dialects per language could result in a serious headache for voice assistants. Headquartered in Montreal, the heart of AI research and development, we are fortunate enough to enjoy a multicultural society that includes many variations of the French language. We put Québécois French against France French to see if the voice assistants could handle it!
For the dialect battle, we asked:
- Can you play ‘All About that Bass’ by Meghan Trainor?
- Is it going to rain today?
- Where is the closest dépanneur?
Winner: Tie!
In one of our finest rounds, Android’s voice search assistant got 2 out of 3 search queries correct. Both performed extremely well in their ‘nearby’ search, using the Québécois-specific word ‘dépanneur’ to find the nearest corner/convenience store within walking distance.
Best (Worst) Answer:
Round 3: International Battle
Our amazing team is made up of superstars from all over the world! We asked a few of them to challenge their voice assistants in their native language to see how advanced voice search was in languages other than the default US English.
For the international battle, we asked:
- Can you play the Game of Thrones theme song?
- Is it going to rain today?
- What is 5 ounces in grams / How many grams are in a kilogram?
Winner: German
Google Home in German took the top prize in the voice search battles showdown! Out of all of the voice searches, Google Home was able to come up with the most in depth and logical answers, and was the only assistant who actually played music without human intervention. We look forward to testing Alexa’s chops, whenever she makes it up to Canada!
Best (Worst) Answer:
When the Searching Gets Tough, the Tough Use Natural Language Processing
Understanding and mastering the nuances of a language is no easy feat – just ask anyone trying to learn a new language with its many complicated rules and exceptions to the rules (I’m looking at you, French!). Even with the incredible complexity of the human mind, it still takes us many years to master our native tongue.
A smart voice assistant must rely on Natural Language Processing, the “automatic manipulation of natural language, like speech and text, by software.” But ‘natural’ is the key here, and with nearly 7000 languages and millions of different accents and dialects steadily evolving around the world, voice assistants are going to need a ton of constantly updating data to keep up.
Big thank you to everyone who participated and all those who shared a few laughs with us! Although in the end, it seems like most of the time the computers aren’t in on the joke – yet.
Image Credits
Featured Image: Unsplash / Hermes Rivera