The MLLP’s candidature to European Parliament call for tenders EP/TRAD/2019/LSTT “Live Speech to Text and Machine Translation Tool for 24 Languages” obtained a strong rating and was ranked sixth, close to the five shortlisted candidatures by large EU and US tech consortia and organizations.
On August 2019, the European Parliament’s Directorate-General for Translation launched the call for tenders EP/TRAD/2019/LSTT “Live Speech to Text and Machine Translation Tool for 24 Languages”. The goal is to develop a tool that is able to automatically transcribe and translate parliamentary multilingual debates in real time covering the 24 official EU languages, in order to provide accessibility for deaf and hard of hearing people as well as to provide a useful service for Members of the European Parliament in accessing debates on screen.
Participation was open to all organizations in the EU (including the UK), and also to organizations worldwide, provided that they have a subsidiary established in the EU (indeed, at least one US big tech company has participated).
Candidates to the call for tenders had to undergo an initial selection procedure to prove their technical and professional capacity to develop the expected tool.
The MLLP research group (VRAIN Institut Valencià d’Investigació Artificial, Universitat Politècnica de València) decided to participate as a single organization supported by our university, as we consider that we have the capacity to develop the expected tool, already having extensive experience and state-of-the-art results in live speech recognition and machine translation in 11 languages.
We are glad to report that the MLLP has indeed passed the selection criteria for technical and professional capacity and all other criteria set by the European Parliament (EP). The results of the selection procedure were communicated by the EP last week, on 20 January 2020.
We were judged on our experience delivering similar services, on the quality of our current automatic transcription and translation systems, on the professional qualifications and experience of our team, and on the quality of our infrastructure and quality assurance means. We fulfilled the requirements set by the EP for each of these aspects separately, with a total score of 30 out of 40 points.
Selected candidates were ranked according to their evaluation in these selection criteria, and the top five candidates were shortlisted to go on to the second step of the procedure. The MLLP’s strong rating has placed us sixth, close to the top five selected candidates, which are all large EU and US tech consortia and organizations, including Microsoft, the Fondazione Bruno Kessler, Sail Labs, Tilde, SDL, the Karslruhe Institute of Technology and AppTek.
This is not the first time that the MLLP has been involved in a EP procedure on speech recognition and machine translation. In January 2017, the EP organized a survey on speech-to-text solutions for their application at the EP, with two phases: an initial evaluation of the solutions available, after which the most promising ones were invited to send a proposal for a proof of concept at the EP. The MLLP was invited by the EP to take part in this survey, and it was one of the organizations selected to contribute with a proposal for a proof of concept at the EP. At the EP’s request, we prepared two demo videos on real-time speech-to-text with automatic language detection, one of them a live demo on an EP session recording, and the other a live speech demo. We invite you to watch these demos of our 2017 technology (which has since seen several improvements):
The MLLP has also recently prepared and released Europarl-ST, a multilingual corpus for speech translation, based on paired audio-text samples for Spoken Language Translation from and into 6 European languages (EN, ES, DE, FR, IT, PT), compiled from EP session recordings. An MLLP research article describing the corpus creation process and our current speech recognition, machine translation and spoken language translation results on the corpus was recently accepted for publication in ICASSP 2020, the world’s largest technical conference on signal processing and its applications.
Even though the MLLP’s candidature will not move to the next round of the EP’s currently active call for tenders, we believe that the MLLP’s positioning in both of these EP procedures is a solid validation of our team’s experience, results and capacity in automatic speech recognition and machine translation. Our automatic transcription and translation systems have been developed in-house, with our own technology, and we have experience in delivering both cloud-based services and on-site installations in numerous technology transfer agreements and contracts (as well as in 3 large EU research projects so far).
We at the MLLP are proud of these results, and we will keep actively looking for new opportunities to continue improving our technology and providing relevant services on speech recognition, machine translation and text-to-speech to EU institutions and other organizations worldwide.