Microsoft has confirmed that its has achieved a new milestone with the company’s speech recognition system which has reached a 5.1 percent error rate and has passed the 5.9 percent word error rate set last year.
The company was successful in reducing the error rate by 12% by using neural net based acoustic and language models. The company researchers had benefited from Microsoft’s Cognitive ToolKit 2.1 and Azure GPUs to explore model architectures to improve speed.
Microsoft is still expected to working on reaching more new levels of accuracy and improve its consumer services like Cortana, Presentation translator and more in the days and years to come.
