What if InterpretBank could automatically prompt you - while you are interpreting - with the translations of the specialized terminology used at the conference, with numbers, abbreviations and places, without the need to perform any search? This is becoming reality now! Through the integration of Automatic Speech Recognition (ASR), InterpretBank will be able boost your interpreting quality, helping you with some difficult aspects of the interpreting process, namely terminology, numbers and proper names.
We have developed this new feature with simultaneous interpreters in mind, but it can be used also for other kinds of interpretation, for example in a Remote Interpreting setting. The incredible advancements in ASR technology that we have seen in the last couple of years have boosted the word rate precision to unprecedented values. Nowadays training is no more needed to reach high precision values, and the systems are getting more tolerant to signal noise, accents, etc. This is starting to make ASR feasible for the use in the interpreting setting. Of course, ASR is still not perfect, and there are situations as it still doesn't work the way we would like, for example with uncommon accents or too many people talking at the same time. However, the quality is expected to keep on increasing in the years to come. InterpretBank with speech-recognition will therefore get better and better.
How does it work: InterpretBank takes the transcription generated by an ASR engine of choice (for example Dragon Naturally Speaking) and use it as an input to find:
- the relevant terminology from your event-glossary: this is one glossary (or more than one) from your database, for example a specialized glossary on medicine or a technical topic you have prepared or received. Intelligent algorithms will show you only the specialized terms of your glossary, no matter singular/plural forms or small deviation from the transcriptions and the glossary form. In some cases InterpretBank will also correct transcription errors. You will get only the terms you may need.
- numbers and unit of measurement: all numbers with related unit of measurement will be spotted and displayed.
- abbreviations: the abbreviations transcribed by the ASR will be spotted and displayed... such as ASR, UN, EU etc.
- name of places: the name of most places, such as cities and countries, will be spotted and displayed.
The extracted information are shown in an ergonomic way. This allows interpreters to immediately spot the needed information, whenever necessary. As interpreting is a very complex task involving a lot of cognitive activities, we have designed the user interface to be distraction-free and clean. Nothing should distract you from your main activity - interpreting between the languages.
This technology has the potential to change the way information is retrieved during interpretation in highly specialized settings. However, the integration of InterpretBank with ASR engines is still an experimental feature, as we have pointed out above the quality of the feature depends not only on the InterpretBank's algorithms, but also on the ASR engine used. We have experimented it with Dragon Naturally Speaking and with Microsoft Bing Speech Servicess. The tests we have performed show a quality of almost 98% in information retrieval if the speech is pronounced naturally with a clear accent. The quality of terminology identification improves if the ASR is able to import the user terminology.
Even if InterpretBank already offers an open interface to integrate the ASR engine of choice directly, for the moment the solution is indicated for organisations and institutions with a dedicated IT infrastructure. However, we are testing hardware and software to allow also freelance interpreter to easily integrate it within their installation.