Microsoft hits a new milestone with its speech recognition system

Microsoft's conversational speech recognition system has hit a new milestone with a 5.1 percent error rate, improved from 5.9 percent reached last year on the Switchboard conversational speech recognition task.

Switchboard is a corpus of recorded telephone conversations that the speech research community has used for more than 20 years to benchmark speech recognition systems.  The task involves transcribing conversations between strangers discussing topics such as sports and politics.

The new study was performed by a group of researchers at Microsoft AI and Research with the goal of achieving the same level of accuracy as a group of human transcribers who can listen to what they were transcribing several times, access its conversational context and work with other transcribers.

Overall, researchers from the latest study reduced the error rate by about 12 percent compared to last year’s findings by improving the neural net-based acoustic and language models of Microsoft’s speech recognition system. Microsoft says that it was able to strengthen the recognizer’s language model by using the entire history of a dialog session to predict what is likely to come next, effectively allowing the model to adapt to the topic and local context of a conversation.

Microsoft’s speech recognition system is used in services like Cortana, Presentation Translator and Microsoft Cognitive Services.