At Leinhäuser, our attitude toward artificial intelligence is a measured one: We aren’t under any illusion that language professionals are about to be replaced by machines, but we’re happy to concede that, in some steps along the way, in certain projects, AI has the potential to make strategic sense.
For many years now, our audiovisual team has been managing voice-over projects for our clients, and in that time, we’ve witnessed some hefty industry transitions. A few years ago, for instance, recordings always took place on site in a studio, whereas these days, more often than not, they’re carried out remotely. And right now, we’re noticing how AI is making in-roads into the world of voice-over. This observation recently crystalized in a request we received from our long-time client Saia Burgess Controls. In the past, we’ve provided them with translation, copywriting, editing, DTP, subtitling, and human voice-over projects. This time, however, they wanted to try out AI voice-over for a series of videos – and we were eager to explore this new tech with them.
Sweet spot: linguistic expertise and technical know-how
As professional linguists, we’re keenly aware of AI’s creative deficits. We know that, these days, a voice-over can be translated and commissioned in a matter of minutes using AI alone – if you and your audience can tolerate translation blunders and robotic voices. But if what you’re looking for is a cost-effective AND linguistically accurate solution for voice-over translation, then you need the full package: linguistic expertise and technical know-how.
Our task
Saia Burgess Controls was looking for a cost-effective AI voice-over solution for 16 training videos with a total run time of 62 minutes. The videos were originally recorded in English and needed to be translated for voice-over into German, Spanish, French, and Italian to accommodate the project partners’ native languages. Since the training videos didn’t contain any sensitive information (for more on AI and information security, read here), we were comfortable recommending that our client opt for AI voice-over with a human touch.
Our take
We took an expert-in-the-lead approach (XITL/EITL). Maintaining an open mind, we first let a state-of-the-art, AI-powered tool take care of the preliminaries: transcribing the original English audio track and providing a first draft translation in each of the target languages. We examined the output and, truthfully, weren’t all too surprised when we discovered that the AI-generated scripts left a lot to be desired …
You see, when it comes to professional multilingual and written communication services, whether content writing, editing, subtitling, or in this case, translation for voice-over, there is always more than meets the eye. In their work, expert linguists make countless conscious and subconscious choices along the way; choices that, for LLMs, are frankly still too abstract. As the author Ted Chiang states in his article “Why A.I. Isn’t Going to Make Art” for The New Yorker:
No wonder, then, that the AI-generated scripts didn’t make the final cut. Without going too deeply into the details (we know that not everybody loves grammar as much as we do and that’s okay …), allow us to illuminate a few of the more overt things that AI fell short on:
- Terminology: As mentioned above, we were working on videos belonging to the same series, so the videos contained a lot of the same terminology. We found that AI didn’t perform well on consistency at all, with terms translated in various ways across videos. It also didn’t consistently translate the introduction of each video … even though each video had the same one.
- Readability: Translation for voice-over is as much about the written word as the spoken word, of course, and the text’s readability needs to be taken into account. AI appeared to gloss over this fact and didn’t seem to mind overdoing it on the alliteration, producing phrasings that it was later prone to trip up on.
- Utterance length: The original AI texts wound up with some major synchronization issues. In technical terms, translation for voice-over is known as a kind of “subordinate translation” because it’s subject to constraints beyond linguistic equivalence, such as timing. When you consider that texts translated from English into German tend to be up to 40% longer, it’s really important to be concise.
When editing the scripts, our linguists took care of the points listed above and a whole lot more to create scripts that were accurate, in-sync, and readable, and ready to be brought to life by AI.
Success story
The final result? Success. Because at Leinhäuser, we offer the best of both worlds: technical know-how and linguistic expertise. Our team keeps up with the latest technological advancements in the language industry, so we’re always best placed to advise you about the implementation of AI in your communications projects, so you don’t have to expend extra time or added costs trialing different tools. Plus, you can always be assured of linguistic integrity – which isn’t always a guarantee with AI in the mix.
In this project, the expertise of skilled linguists proved, once again, to be indispensable and inimitable. And, despite the hype, it’s unlikely that we’ll be able to bypass the human element in language and communications projects entirely in the near future – language is simply all too human. If speed, cost-effectiveness, and scalability are your top priorities, AI voice-over with an expert-in-the-lead approach could suit you. But to foster genuine connection and authenticity? We’d keep all elements human.
If you’re looking for experts to lead your AI/human voice-over projects, look no further. Our experienced account managers will collaborate closely with you to fully understand your expectations and develop workflows tailored to your specific requirements.