Speech Technology Magazine

 

Speech Translation Helps Multilingual Conferences Express Themselves

New technology means that speech translation is quicker, more accurate, and cheaper than ever—and international conference goers will reap the benefits
By Sue Ellen Reager - Posted Jul 23, 2018
Page1 of 1
Bookmark and Share

Conferences involving multilingual participants have often proved challenging, and in the past not all speech translation attempts have succeeded, for the usual reasons: unclear speech, mediocre machine translation, poor audio equipment. Speech-to-text translation has often been hurt by speakers including too many jokes, imagery, and metaphors, all of which confuse translation engines.

But the advent of “neural translation,” which relies on a large artificial neural network to predict the likelihood of a sequence of words, is vastly improving the quality of speech translation. Neural translation, when combined with artificial intelligence and machine learning, can lead to speech translation that’s humanlike in its quality, and companies like DeepL have begun to produce the most accurate machine translations on the planet, with natural word flow, good grammar, and fewer errors than other translation engines. The addition of context engines like Tywi guide the translation software to produce more contextually relevant translations, with improved word choice.

The quality of the audio equipment also plays a major role in speech translation success in a conference setting. The insurance company Nationwide used speech translation at a recent conference to effective ends (more about this below), and its foray into speech translation was cleverly improved by using a $2,200 microphone as well as engaging speakers who were accustomed to giving public presentations. Indeed, it is astonishing how much difference a great (expensive) microphone makes toward accurate speech recognition.

So there are ways to create an atmosphere conducive to effective and understandable speech translation in real time. And the best news is that the speech recognition engines themselves are being developed and improved every day, marching forward to more perfect understanding of imperfect speech. Below are three examples of companies and conferences that have pulled off stellar speech translation in a live setting.

Nationwide ran a very effective experiment during a recent intense two-day seminar in Las Vegas. There were 30 speakers over two days, each speaking approximately 20 to 30 minutes, and their microphones were fed through translation technology to generate instant captions followed by translation into subtitles in a dozen languages. Those subtitles were also available as text-to-speech (TTS) so that the audience could choose to listen rather than read. The results were surprisingly understandable and usable, representing a huge leap forward in conference globalization.

Honda Europe routinely faces translation challenges (imagine up to 13 languages spoken in one room), and it also overcomes them using speech translation technology. At a recent management meeting the company featured multiple speakers, and each speaker hailed from a different country, and thus had a markedly different accent. Honda overcame these barriers by engaging a “parrot,” a person who repeated into translation software everything said by the person behind the podium. The translation program recognized the parrot’s speech and then generated and displayed subtitles in multiple languages on side screens during the meeting. 

Finally, a major Asia-Pacific conference that has been held for decades in English only, with periodic interpretation only in the keynote hall, revolutionized its programming by translating all presentations in all rooms simultaneously. Human translation on this scale has never been affordable, once you factored in not only the cost of interpreters but also booths, cables, transmitters, receivers, and headphones for a large audience. The cost of traditional human interpretation over the three-day conference added up to $180,000 per language—just for the interpreters. Multiply $180,000 times five languages at the conference and you arrive at a tidy sum. Importantly, that price tag assumes the use of a “no-equipment solution” that removes the need for booths, cables, transmitters, receivers, and headphones—which could cost as much per language as interpreters, and double the total. The use of speech translation technology, then, converted a potential cost of nearly $2 million down to a few thousand per language.

Soon, expect to see speech translation as a simple flip of a switch in major convention centers. These halls will already be wired and ready to translate. And speech translation will be an expected service for conferences large and small. 

Sue Reager is president of Translate Your World, developers of software for across-language speech communication, and can be reached at sreager@internationalservices.com.

Page1 of 1