Many companies are promising the world to their clients while leaning heavily on technology, especially recent advances in Artificial Intelligence. When it comes to video/audio localisation, the below image comparing workflows presents a strong case for adopting AI solutions over traditional methods, and a surface level web search for such services will back this up.
But do the results back up this promise? If only it were so simple.
We decided to put current AI video localisation tools to the test. Keep reading to see what we discovered:
Background
Having a slick video introducing your company is a no-brainer. You can take decades of growth and success and distill them into a 1-2 minute segment. With some preparation and marketing savvy, you’ll have an incredibly valuable piece of media that can increase sales, aid recruitment, and become a linchpin in your company’s identity moving forward.
Significant effort and resources go into producing this kind of video. The larger the company, the greater the effort. Additionally, the wider a company’s reach, the more languages the video should be translated into. So, here’s the challenge: Can a company profile video be translated into multiple languages without being “re-made” for each one? What is the most efficient way to complete the localisation process? Is AI the magic fairy dust we’ve been waiting for to cut costs and save time?
The challenge
Translate and dub Prestige Network’s video into 4 languages (Greek, Polish, Chinese, and Arabic) pushing AI tools to their current limits.
The results
There is indeed a process and pathway to video localisation using AI, but it’s not the freeway one might expect. The processes of transcription, voice isolation and removal, translation, voice cloning, dubbing, and editing leave plenty of room for error.
Even with our technicians’ best efforts, a completely accurate translation was not achieved. We resisted the temptation to reach out to our many linguists, as this project was all about testing AI as an all-in-one solution.
Despite not being completely accurate, the results are still quite astounding. At just a fraction of a typical video localisation budget, we have 4 videos in which the speaker (Tim, Head of Technology) maintains his voice and cadence in these 5 languages, speaking what would be considered “acceptable” language. He would be understood by a native speaker of the target language, but the translation would not be 100% accurate at all times.
So, is there a place for this technology and type of video localisation process? Yes, in situations where budgets are limited and margins of error are not critical. There are scenarios where being understood in another language is the main goal, rather than achieving a complete and perfect translation of the source material.
We found that AI alone has very limited application in a localisation context. However, when paired with our in-house technologies and technicians, and supported by linguistic expertise, it can deliver high-quality results at a reduced cost to our clients.
