Skip to main content
European Commission logo
Maltese
Education
Knowledge Centre on Interpretation

Speech recognition

How can speech recognition affect our work?

Blog

Speech recognition and speech-to-text solutions have made significant progress over the past few years. They can further the development of a range of applications benefiting everyone, e.g. terminology extraction, text summarisation and text analytics. Meeting participants and conference interpreters create multilingual content that can help fine-tune such tools. How can the interpretation community, research and industry initiatives reap the benefits of our multilingual environment?

Language
English

I am writing my master thesis on this topic with the aim to examine whether the speech-recognition software could be of help in consecutive interpreting. I will be using the Gile´s effort model (1995) to see...

I am writing my master thesis on this topic with the aim to examine whether the speech-recognition software could be of help in consecutive interpreting. I will be using the Gile´s effort model (1995) to see if a speech-recognized source text could facilitate efforts (such as listening and analyzing (L) and remembering (Rem)). This method would be similar to SimCons - interpretation with a SmartPen, only this time a source text would be written, which could introduce a new method - the SightCons (a fusion of Sight translation and Consecutive Interpreting). I will be doing an experiment with two groups of advanced conference-interpreting students, in which they will be doing two consecutive interpretations - a "classical" consecutive with note-taking, and "SightCons" with the help of the Dragon NaturallySpeaking software and note-taking (if they wish). Both interpretations will be rated upon quality criteria (Kurz 2001). I would really appreciate your ideas and comments regarding my experiment and thesis in general :) Also, once the experiment has been conducted, I would be more than happy to share the results here in the group.

Read more

Language
English

Hi Sandra, A few initial questions for your research design: - Will interpreters receive a perfect transcription of the speech, or the version generated by Dragon? There might be differences in how that affects their work...

Hi Sandra, A few initial questions for your research design: - Will interpreters receive a perfect transcription of the speech, or the version generated by Dragon? There might be differences in how that affects their work and whether they want to receive a transcription or not. - Will the transcription be word-for-word, or will you simply be providing interpreters with the "difficult" parts of the speech - names, numbers, named entities, etc.? I'm particularly interested in SightConsec right now, and look forward to seeing the results of your research! When do you expect they will be out? Best, Josh

Read more

Language
English

Hi Joshua, Thank you so much for your remarks and enthusiasm for my thesis! In my research, I also stumbled across your thesis (Goldsmith, J. & Holley, J. (2015).“Consecutive Interpreting 2.0:The tablet interpreting experience.“Unpublished MA thesis...

Hi Joshua, Thank you so much for your remarks and enthusiasm for my thesis! In my research, I also stumbled across your thesis (Goldsmith, J. & Holley, J. (2015).“Consecutive Interpreting 2.0:The tablet interpreting experience.“Unpublished MA thesis, University of Geneva.) which goes in the similar direction with my reflections. There is also another MA thesis, published by Aneta Rafajlovska in which she gives a conceptual design proposal for a computer assisted tool for the transcription of symbols into text, which goes even further into the future. 1. I actually wanted to prepare a transcription of source speeches manually (without using a software), but then my supervisor (Prof. Dr. Mira Kadric) suggested I should do it with a program. I am expecting to have an accuracy rate of at least 95.5 % after using the software for 17 days and by choosing English as source language. 2. The transcriptions (for both speeches) should be 95.5% accurate, including all parts of the text. Furthermore, there is a possibility to highlight some of the information- such as names, numbers, numeration, etc. It would also be possible to separate clearly text paragraphs by programing the software to recognize pauses longer than xy minutes as a new paragraph. This would help an interpreter to have a clear structure when comparing the notes with the "transcription" of the source text. I am hoping to conduct the experiment before Christmas holidays and to process the results by end of January. However, if you would like to have some further details on the current status of my research, I would be more than happy to send you some of my documentation (in German). Regards, Sandra

Read more

Language
English

Dear Sandra,1. SimConsec: I have been using Notability with the dictation function enabled for most non-confidential meetings for several years. With an Airpod discreetly lodged in one ear it works great for me in a professional...

Dear Sandra,1. SimConsec: I have been using Notability with the dictation function enabled for most non-confidential meetings for several years. With an Airpod discreetly lodged in one ear it works great for me in a professional setting. 2. Likewise, I have found speech recognition of my *output* useful as a memory support tool during Sim. Dragon Dictate is, however, trained to transcribe my voice. I have had Brähler make me a cable to transfer the speaker audio to the microphone input of my MacBook for tests with ASR: the transcription with Dragon, Cortana (Microsoft), Google and Siri have been pretty useless in reverberant conference halls. Claudio Fantinuoli and the Hansen-Schirra Team @ FTSK Germersheim have however been using tailor-made algorithms and Interpretbank does a fine job with it.In my book "Jenseits simultanfähiger Terminologiesysteme", the proposed workaround was to train the system to at least scroll your chronological glossary with the speaker. Detecting the limited number of phonemes likely to match the glossary entries has worked well enough for me. The subject however is a complex one: ASR infringes speakers' copyright unless authorised beforehand. This is why we have set up a controllable yet real conference in 2015 to test CAI tools under controllable conditions: ShowCase Live conferences series (heicic.de). Next one is coming up on July 3rd and you are most cordially invited. ChristophBook on CAI: http://independent.academia.edu/StollChristophp.339: In Ergänzung des Pull-Prizips mit Eingabe der gesuchten Benennung über die Tastatur wird parallel ein Push-Prinzip verfolgt: Das System analysiert auf Grund der von der Computerlinguistik zu formulierenden Algorithmen, welche die jeweils wahrscheinlichen Dolmetschstrategien abbilden, ständig den Ausgangstext unter anderem auf Realia und nicht in der Datenbank vorhandene Benennungen, recherchiert diese im Hintergrund und bietet die Ergebnisse zur Selektion an.Taucht im Redefluss die Benennung Hexansäure auf, würde das Spracherkennungssystem nach Elimination der Gemeinsprache über eine Stopp-Liste (Vortrag F. Austermühl 21.01.2009 am SUED Heidelberg), die Fachtermini im Hintergrund nachschlagen und in einer fortlaufenden Liste mitscrollen. Neben vielen nicht benötigten Äquivalenzen könnte der Dolmetscher aus der laufenden Liste das Benennungspaar Hexansäure - hexanoic acid selektieren und in das Translat integrieren.Ist kein AS-ZS-Benennungspaar auffindbar sollte das System analog zu den nicht CAI-gestützten Wortfindungsstrategien gleichzeitig in verschiedene Fenstern eskalieren und Angebote zur Unterstützung der Kompensationsstrategien darstellen:

Read more

Language
English

It sounds a very interesting project! We definitely need more research on whether these new tools can improve interpreters' work or add a layer of complexity/distraction. I'll be happy to invite you to present the results...

It sounds a very interesting project! We definitely need more research on whether these new tools can improve interpreters' work or add a layer of complexity/distraction. I'll be happy to invite you to present the results of your research to the "Task-Force on digital transformation". Let's keep in touch!

Read more

Language
English

Thank you Marc! I will let you know, as soon as I kick off with the experiment and get some outcomes.

Language
English

This is a demo of InterpretBank showing how voice recognition, tied to pre-prepared glossaries can help simultaneous interpreters. The software recognises terms spoken if they are already in your glossary and shows the translation from the...

This is a demo of InterpretBank showing how voice recognition, tied to pre-prepared glossaries can help simultaneous interpreters. The software recognises terms spoken if they are already in your glossary and shows the translation from the glossary on screen. (It also does numbers always). If it can be made reliable it will be brilliant. https://youtu.be/w5dC17o0Sts

Read more

Language
English

I think this question (the last sentence) is too vague and relates too little to the intro about speech recognition

Language
English

Thanks for this comment and sorry for not having reacted earlier. In fact the texts that have been posted in some of the initial communities of the KCI are there only to awake curiosity and interest...

Thanks for this comment and sorry for not having reacted earlier. In fact the texts that have been posted in some of the initial communities of the KCI are there only to awake curiosity and interest and invite people to participate. In this case the texts is trying to make reference to the fact that we are particularly well placed to contribute to this debate because we work in a multilingual environment by definition and that we should embark on that debate. Probably you are completely right and it is not sufficiently clear. All your suggestions for improvement are more than welcome.

Read more

Files

Wiki docs

No content available.

Members

Contact selected
No members available.