This insight shaped the development of elysium® Live Translation 2.0.
The new version is not the result of theoretical concepts or internal testing alone. It is the outcome of a full year of real-world use—during guided tours, pilot projects with major cruise lines, and day-to-day operation with countless groups.
Throughout this process, we closely observed where automated translation already performs reliably—and where a technically correct word-for-word translation still falls short of delivering a truly good translation.
Live Translation 2.0 combines these practical insights with a fundamentally enhanced approach to processing language, context, and pronunciation.
Developed in the Real World, Not in the Lab
With Live Translation 1.0, our goal was to understand how AI-powered translation performs in everyday group communication.
Not in a quiet meeting room with carefully prepared example sentences, but under real conditions: with different guides, various accents, background noise, lengthy explanations, technical terminology, and back-to-back tours.
Several major cruise lines tested our Live Translation solution during pilot projects on board their ships. Another pilot with a large cruise operator will begin in September. In addition, a company in Switzerland has been using the solution on a daily basis for several months—serving around 30 groups per day.
This intensive real-world usage has been invaluable for development. Only continuous daily operation reveals whether a solution remains reliable when it is no longer demonstrated occasionally, but used every day under changing conditions.
Feedback from these deployments directly shaped Live Translation 2.0. Many of today's features are based on situations that would have been almost impossible to anticipate from behind a desk.
Why Word-for-Word Translation Isn't Enough
Spoken language is rarely as structured as written text.
People begin a sentence, change their wording while speaking, repeat themselves, or add another thought before finishing. In German, the main verb often appears only at the very end of a long sentence.
A translation that processes every word immediately and in isolation can therefore easily miss the speaker's actual meaning.
Live Translation 2.0 deliberately moves away from this rigid one-to-one approach. Instead, the AI processes larger linguistic units before translating them, allowing it to better understand sentence structure, meaning, and context.
This results in an initial latency of approximately five to ten seconds. Rather than being a technical disadvantage, this short delay is an essential part of delivering high-quality translations.
Human interpreters also need to understand what a speaker intends before translating. A professional translation should never guess how an unfinished sentence is likely to end.
Our goal is therefore not the lowest possible latency, but the best balance between speed, clarity, and translation accuracy.
Intelligent Text Optimization—Without Summarizing the Content
Before translating spoken language, Live Translation 2.0 intelligently refines the recognized text.
The AI removes meaningless filler words, detects accidental repetitions, and splits very long sentences into shorter, easier-to-understand ones. Self-corrections made by the speaker also don't necessarily have to appear twice in the translated output.
Here's a simple example: A guide starts explaining something, interrupts themselves, and then restates the idea more clearly. A strict one-to-one translation would translate and read out both versions, creating unnecessary repetition for listeners.
Live Translation 2.0 recognizes these speech patterns and delivers the complete message in a clearer, more natural form.
The content itself is never summarized. The AI does not decide which information is important or unimportant. It does not omit relevant details or create a shortened interpretation.
Only the linguistic form is optimized:
- meaningless filler words are removed,
- accidental repetitions are eliminated,
- long sentence structures become easier to understand, and
- the complete information remains intact.
The result is not only easier to translate—it also sounds significantly more natural when spoken aloud in the selected target language.
Context and Pronunciation Make the Difference
The same word can have different meanings depending on the situation. A generally correct translation is not always the technically correct one.
At the beginning of a tour, Live Translation 2.0 analyzes the first spoken sentences. Based on this input, the AI gradually recognizes the environment in which it is being used.
Is this a sightseeing tour on Madeira? A factory tour? Or are highly specialized steel construction terms being used?
With every sentence, the system gains additional context. As a result, translations evolve from general language processing to increasingly precise, topic-specific translations.
Speech synthesis is equally important. Even a perfectly translated sentence loses value if pronunciation, intonation, or speech rhythm sound unnatural.
Our real-world testing has shown—especially for Mandarin and other Asian languages—that intelligibility depends heavily on the quality of the text-to-speech engine. A technology that performs exceptionally well in one European language is not automatically the best choice for every language.
That's why Live Translation 2.0 uses language-specific optimized components. Speech recognition, translation, and text-to-speech are combined differently depending on the requirements of each language.
The technical complexity remains invisible to users. What matters is that the translation sounds clear, natural, and easy to understand.
What's New in Live Translation 2.0
Speakers no longer need to select their input language before starting. They can begin their tour, training session, or presentation immediately, while the AI automatically detects the spoken language.
The best results are still achieved when speakers use their native language. This allows them to speak naturally, use confident sentence structures, and focus entirely on their audience instead of the technology.
Automatic language detection is not limited to the currently available target languages. Listeners can currently choose from 29 target languages, with additional languages available upon request.
Based on real-world usage data collected so far, our processing pipeline—including speech recognition, text processing, and translation—currently achieves 99.6% correctly processed content.
This figure is not based solely on laboratory test sentences. It reflects a wide range of speakers, topics, speaking styles, and real operating conditions.
Nevertheless, Live Translation remains an AI-powered system. Poor internet connectivity, low audio quality, unclear pronunciation, or highly specialized terminology may still affect translation quality.
For legally or medically critical situations, Live Translation is not intended to replace a certified or otherwise qualified human interpreter. However, for guided tours, training sessions, events, and other forms of multilingual group communication, continuous real-world operation has demonstrated just how capable automated live translation has become.
From Tourism to Every Industry
The first applications of elysium® Live Translation were primarily in tourism: city tours, shore excursions, cruises, and international travel groups.
Over the past year, however, it has become clear that the same communication challenges exist across many other industries.
International employees need to understand safety briefings. Visitors want to follow a factory tour. Training sessions should not have to be repeated for different language groups. Conference attendees expect to receive the same information in their own language.
Live Translation 2.0 is therefore ideally suited for a wide range of applications, including:
- factory and company tours,
- training sessions and safety briefings,
- onboarding international employees,
- conferences and corporate events,
- museums and educational institutions,
- international delegations and visitor groups, and
- guided tours and shore excursions.
Listeners simply scan a QR code with their smartphone and open the session directly in their web browser, where they can select their preferred language. No app installation, user account, or registration is required.
Alternatively, Live Translation can be integrated with compatible tour guide systems, allowing smartphones and traditional receivers to be used together during the same event—or enabling organizations to continue using their existing hardware.
Live Translation 2.0 is no longer just a translation solution for tourism. It is designed for any company or organization that needs to communicate reliably with multilingual audiences.
For us, Live Translation 2.0 marks an important milestone—but not the finish line. The AI is continuously evaluated, refined, and improved based on new insights gained from real-world deployments.
Pilot projects will continue to play a central role in this process. They don't show us how translation performs under perfect laboratory conditions—they reveal what it needs to achieve in everyday practice.
Ultimately, everyone who uses Live Translation benefits from this continuous improvement.
