Things have been moving fast here at IV to bring key improvements to our customers, particularly in the Capital Markets space.
There are two key areas that our clients focus on: Training the system to deal with their complicated language needs, and identifying which language their users are speaking.
V5.2 brings access to more of our new speech recognition networks that give our highest ever accuracy, even across different accents. However, these networks are harder to adapt for new words and phrases than their predecessors as we move away from the world of “phonetic” speech recognition towards end-to-end audio to text networks.
V5.2 also brings a new training methodology that overcomes these issues to give high “Conversational” accuracy, as well as the ability to get high precision for user-trained words and phrases. This is the next stage along our journey to a truly self-learning speech recognition, which we aim to release next year.
Language identification is a major issue where you have a multi-lingual workforce, who can change language mid-sentence. The new release contains a major update to allow better identification of different languages, especially across multi-channel files.
Other improvements are faster diarization, especially for longer files, with an accuracy tweak as well.
There are a number of other “under the hood” improvements: Full details here.
Intelligent Voice offers multiple deployment options from fully on-prem, through private cloud to a secure private SaaS service.
Existing users can download the new release from their support portal, but please feel free to get in touch with your support representative to talk more about deployment options, or contact email@example.com
RelativityOne user? Take advantage of our native processing and review options.