Researchers at
Microsoft have released details of new speech recognition technology that they say translates conversational speech as well as a human. "We've achieved human parity," says Microsoft's chief speech researcher
Xuedong Huang in an announcement. "This is an historic achievement."
For a considerable length of time science fiction movies have been envisioning the future as one where humans talk to machines pretty much like family and friends. In reality, be that as it may, using voice to interact with machines has been maddeningly frustrating, with Siri frequently mixing up "open up my email" for "look into some fail," for instance. This is evolving.
A group at Microsoft's
Artificial Intelligence and Research group have built a
voice recognition system that commits errors at an indistinguishable low rate when compared to professional human transcribers. The researchers investigated an assortment of recurrent neural networks and had the most success with a LSTM (Long Short-Term Memory) model. The model was trained with a 30k-word vocabulary for 2,000 hours on a server farm comprising of multiple GPUs.
Microsoft plans to utilize the technology in
Cortana, its personal voice assistant for Windows and the Xbox One, and also speech-to-text transcription software
The group's next objective is to enhance the engine's robustness so it can be used as a part of real-life situations, for example, on crowded city streets or while driving. They additionally would like to inevitably motivate it to work with multiple users simultaneously.
