NCSpeech upgrades voice AI for more accurate live speech

NCSpeech, a participant of Astana Hub, has improved its voice AI so that it recognizes live speech more accurately. The idea is for the machine to understand people the way they actually talk, not the way they read from a script. The news was reported by Astana Hub.

Live speech is full of pauses, interruptions, slips and shifting pace, and that is exactly where recognition systems make most of their mistakes. A single error in a transcript can distort an amount, a name or a deadline. The more accurate the text produced from voice, the safer it is to build further work on it: meeting minutes, customer requests, spoken instructions.

Estimate how much working information in your company lives in voice messages, calls and verbal agreements, and how much of it gets lost afterwards. Before adopting speech recognition, test it on real recordings from your staff and customers: with background noise, fast speech and the mix of languages you actually use. The better machines understand ordinary speech, the easier it is to ask in plain words: in 1Cifer, a role-based agent answers from 1C, CRM and bank data in Russian and Kazakh.