From business meetings to banter, Google’s new speech translator has its strengths and weaknesses, says Oliver Barham at Locaria. But does this spell the end of learning foreign languages?
Barham and his colleagues put Google’s English-Spanish real-time speech translator to the test
Testing the boundaries of technology is always fun, especially when that technology promises to dissolve one of humanity’s oldest barriers: language. When Google unveiled its real-time English-Spanish speech translator in May, powered by Gemini AI and available to AI Pro subscribers, we at Locaria couldn’t resist the urge to put it through its paces. After all, what’s more fun for a bunch of linguists and localization experts than trying to trip up a machine with idioms, rapid-fire banter, and some serious business jargon?
Let’s start with the big picture. Google’s translator is a step towards a world where language differences don’t slow down global business, international teamwork, or even a casual chat between friends on different continents. In structured settings (think business meetings, technical presentations, or corporate updates), the tool is already surprisingly capable. It preserves the core data, gets the numbers right, and mostly keeps the tone professional. For global teams, this is a game-changer: suddenly, a meeting with colleagues in Bogotá, Madrid, and London can flow with far fewer awkward silences and “Sorry, can you repeat that?” moments.
Hannes Ben, CEO at Locaria and polyglot, had another perspective to share after he first read the report: "from a linguist’s perspective, the long-term potential of Google’s real-time translation technology is quite exciting, especially for languages that never really get a simultaneous or consecutive interpretation in the majority of international conferences. Mostly, you only get English, Spanish, French, Chinese, and Arabic fully covered. Imagine a future where speakers of endangered or regional languages can participate fully in international meetings or share their culture with the world, simply by speaking their mother tongue instead of focusing on how to say something correctly.”
But let’s not pretend we didn’t have some fun trying to break it. We threw everything at it: casual greetings, technical jargon, rapid back-and-forth. Sometimes, the results were genuinely impressive. A business update about response times and efficiency gains came through almost flawlessly: “Response times fell 12%” is exactly what you want to hear in a boardroom, and the Spanish version was just as clear.
Other times, the AI’s earnest literalism gave us some interesting results: “Me levanté con el pie izquierdo” (I woke up on the wrong side of the bed) became “I woke up with my left foot”, a phrase that, while anatomically accurate, doesn’t quite capture the spirit of a bad morning. “Derramé el café encima de la camisa” (I spilled coffee on my shirt) transformed into “I broke my coffee on my shirt.”
Our structured evaluation (see the full report on Locaria’s website for the detail) paints a clear picture. In the business context, Google’s tool scores in a respectable five to seven (out of 10) range for both languages. But when things get informal, idiomatic, or fast-paced, the cracks show. Scores for casual and idiomatic speech drop to the three to four range. Naturalness and fluency, especially from English to Spanish, can plummet as low as 2/10, meaning the output is often robotic, stilted, or just plain odd.
There’s also the issue of timing and delivery. In real conversations, we interrupt, overlap, and finish each other’s sentences. The AI? Not so much. Once you start speaking, it takes a few seconds before kicking in, so there’s a very noticeable delay. For now, that means it’s best suited for turn-based, structured dialogue, not the lively, overlapping chatter of a family dinner or a brainstorming session.
Despite these growing pains, it’s hard not to be excited. Every test, every awkward translation, every moment of “Did it really just say that?” is a step towards something better. The potential is enormous: seamless multilingual meetings, instant comprehension across continents, and a democratization of communication that could reshape business and society.
For Google, and all of us in the industry of multilingual content, the message is clear: keep going. The progress is real, the opportunity is vast, and the fun is just beginning. If the team at Google wants to reach out for further testing, training, and feedback, hit us up. After all, the only way to teach an AI to ‘get’ a joke is to keep talking to it and challenging it.
Hannes left me with one final thought as I discussed the report with him: “if real-time translation becomes universally available and reliable, will learning new languages lose its appeal? Or is it that translation and interpretation become skills left only to the grand masters of language, unicorns of the multilingual realm, able to compete with AI?”. Only the future will tell.
Locaria
Locaria is a pioneering global multilingual content production agency which specialises in supporting inhouse marketing activation, e-commerce content delivery,...
