Instantaneous and accurate machine translation could help journalists report on breaking stories, transform the business models of call centres and telecom carriers and even ease world diplomacy. A EUREKA-funded collaboration has brought this ambitious concept one step closer to reality.
Imagine you are a financial journalist about to follow an important streamed press conference in a language you do not understand. Then imagine you are able to follow every word perfectly, instantaneously. Through pioneering new speech recognition, sentence identification and dynamic vocabularly technologies, the EUREKA network project MEDIATRANSLATOR has developed the first real-time video translator, linguistically optimised for the financial sector.
Instant accurate translation
This success is the result of successful collaboration between Lexifone Communication Systems in Israel and the German company European Media Laboratory GmbH (EML).“The technical challenge here was not so much the issue of translation as much as the issue of speech recognition,” explains project coordinator Dr. Ike Sagie, Managing Director of Lexifone. “Current technology does not handle two people speaking at the same time very well. Our breakthrough was achieving the ability to differentiate between two speakers, increasing the accuracy of the transcription.”
‘We have since identified a number of potential new service offerings and built contacts with new customer communities’
EML provided deep-learning-based speech recognition algorithms and a highly scalable transcription platform for automatic transcription of the speech content of video feeds. Smart segmentation technology supports instant translation, and the speech recognition vocabulary can be dynamically updated within hours. The team believes that the combination of these solutions provides for superb transcription accuracy. Finally, a demonstration application was developed in order to help visualise the results.
Starting a global conversation
Both partners see MEDIATRANSLATOR as an important stepping stone towards their respective global ambitions. “I set a goal ten years ago to develop a solution that could overcome global language barriers and enable communication with anyone on the planet,” explains Dr. Sagie. “The reason why the major tech players have not already done so is because such an innovation requires extensive customisation. There is simply no one-size-fits-all solution for this level of accuracy.”
Other potential beneficiaries fo this ‘industrial strength’ machine translation include high level diplomats involved in delicate calls with foreign coutnerparts, major call centres and telecom carriers. Machine translation could create huge operational efficiencies, and help carve out new revenue streams. “This project represents an important building block towards the realisation of my overall concept,” explains Dr. Sagie. ‘I’m advising large organisations, call centres and telecom carriers how to build customised tailored solutions.”
For EML, the EUREKA project has helped the firm to enrich its technology portfolio with assets that can be re-used in other areas including smart homes or the smart vehicle market. “Apart from the successful R&D activities, we have benefitted business-wise by having been exposed to financial services, a new market segment for us. We have since identified a number of potential new service offerings and built contacts with new customer communities,” says Volker Fischer, Head of Research at EML.
EML received funding from the German central innovation programme for SMEs (ZIM), a funding programme of the Federal Ministry for Economic Affairs and Energy (BMWi), while Lexifone Communication Systems was funded by the the Israel Innovation Authority.
Looking to the future, Dr. Sagie notes that interesting research results are currently being produced on vocalising text in one’s own voice, and that he expects to be able to embed this technology within five years. He also stresses that this is a long term project, one in which a great deal of time and resources must be invested. “A major obstacle is hesitancy because the concept is so new,” he says. ‘People are still unsure about having a robot interpreting their call.’