Zoom Media attended an event hosted by the Dutch
The Dutch government is confronted with quite a challenge when it comes to the use of interpreters. In the current situation interpreters are hired for police interrogations, conversations with immigrants, trials and so forth. Interpreters are needed on location almost every time and the total costs nowadays are around 80 million Euros per year. Together with the public sector, the Dutch government now aims to find innovative ways of cutting costs, work more efficient and make sure it is ready for the (near) future where it expects an ever growing demand for such services due to an increase in immigration.
Zoom Media offers speech recognition solutions in various languages and can build and host high accuracy models at a rapid rate. Now speech recognition does not do what an interpreter does of course, it converts speech into text in the same language, it doesn’t translate. Even so it can be a useful tool within the workflows currently in place. Speech recognition, in various cases, can serve as a starting point. It can lay the groundwork so to speak. And so during the various sessions with different groups, a few interesting innovative ideas came to mind.
Police hearings these days are written out by policemen and -women leaving them with less time to do their actual job which is keeping us safe. Using speech recognition, interrogations can be automatically transcribed leaving only some minor correction work for an interpreter.
Building on the previous example, the transcripts of these interrogations are linked with timestamps to their respective audio files, meaning they are easily searchable as well. This opens up possibilities for trials where there can arise discussion over what a defendant said exactly during his interrogation. Since those interrogations are made searchable, a judge could simply type in a specific word about which there is discussion, look up the exact moment in the audio file and play it out loud in court for all to hear and ending the discussion.
Last but not least, although it isn’t perfect, speech recognition results could be automatically transcribed using a translation engine. Now the results are way too literal to be regarded as an exact translation, but again it can serve as the groundwork for others to rewrite. It can serve as a basis for interpreters. Plus, for people within an organization who are only concerned about what the broad outline of a conversation was, the literal translation can be enough. Also, same as with the trials, one can search through the audio thanks to the timestamps added