Among some people in operations, anything related to artificial intelligence (AI) often is met with trepidation because of the potential of AI-based systems eliminating positions in Ops. This fear is somewhat justified even though there is mounting evidence that AI can help financial services firms replace tedious tasks — and not staff members — with systems.
One such tedious task is the transcription of voice recordings, which even AI has challenges doing with high rates of accuracy, which is essential for regulatory reporting. Yet one vendor is saying that there is hope that AI can improve automatic speech transcription, an area that has been the focus of GreenKey Technologies, a Chicago-based maker of what it calls “an AI and voice-driven collaboration tool for financial market participants.”
In fact, GreenKey reports that the U.S. Patent and Trademark Office has accepted the vendor’s application for improved speed and transcription accuracy of speech transcription “based on multiple speech-to-text engines,” officials say.
The vendor is arguing that its speech recognition engine, GreenKey Scribe, can “reduce transcription error rates by 50 percent using the approach documented in the new patent. “
The Scribe offering exploits “an array of transcription engines trained on different content domains and automatically selects the engine that will perform best on a particular segment of a conversation,” according to the vendor. “The transcription engines that Scribe leverages can be a combination of models built by GreenKey, engines built by partner transcription providers and cloud transcription services.”
The vendor also argues that its way is better than “multi-engine approaches” that distribute audio to several engines in parallel and the compare the transcriptions. “This process can be slow and computationally expensive,” say GreenKey officials.
By contrast, the GreenKey method via the “Scribe SwitchBoard,” has the capability to “analyze an audio file in less than a second to determine which engine will transcribe it best, without having to send it to all of them,” vendor officials say.
In addition, GreenKey Scribe’s design helps end-users to transcribe “noisy, jargon-heavy audio with high accuracy in a variety of industries,” including financial services. The system also transcribes voice recording that have “a variety of accents, languages and audio quality environments,” officials say.
Scribe’s SwitchBoard is a new approach to “how we train AI to transcribe human speech,” says Dr. Tejas Shastry, chief data scientist for GreenKey Technologies, in a prepared statement. “Firms can train their own customized models, and Scribe can leverage them all together. Beyond increasing accuracy, it opens up the possibility to transcribe multi-language conversations with a single engine.”
So, if the GreenKey offering pans out, this AI-based innovation (and other like it) might make life easier for Ops and compliance staffs that must know exactly what was said regarding securities transactions. If nothing else, it may make enormously easier the need to review transcriptions for small errors. It might go a long way toward ending painful and tedious transcriptions done by humans. AI-based systems may also uncover problems that human transcribers might have missed — all without threatening anyone’s job.