That process has involved building technology on the back of text-based translation and speech recognition platforms like Cortana and Bing Translator that Microsoft had already developed, as well as returning to a technology known as “deep neural networks,” or DNN, that was pushed aside several years ago as researchers found other method of machine translation more promising, says Vikram Dendi, strategy director for Microsoft Research. “The whole process for us in research has been illuminating.” “The way people interact turns out to be really interesting,” says Peter Lee, corporate vice president at Microsoft Research.
#Skype translate bot software
Errors that the software makes today will disappear as the software logs more examples of natural human language, of the way humans write and speak differently, and of how they word things differently for social media, e-mail, chat, and spoken conversation. The more users speak through Microsoft’s translation platform, the better it understands human language and the more accurate it becomes. The latter aspect is the key to Skype Translator’s future success. But the fact that Skype Translator is right most of the time-so often that not only can the average person correct for errors using conversational context, but also often enough for Microsoft to feel comfortable releasing it as a consumer product-represents a leap forward for machine translation specifically and machine learning more generally. The software still sometimes gets hung up on idiom, nuance, on problems created by tone, common mispronunciation, or the lazy way most of us enunciate the words of our mother tongues.
For the user, the experience is much more simple: A mere half-second after one party finishes speaking, an audio translation of his or her speech plays through to the recipient alongside a text transcript that runs alongside the video or voice call. All of that complexity happens in the background thanks to a tremendous amount of computing power and software wizardry that the bot can access in the cloud. Between speakers, a third participant on the call-a Skype Translator bot-processes each speaker’s language through a multi-tiered, cloud-based application for speech recognition, translation, and synthesis. (Microsoft Research has already demonstrated technology that can translate in the user’s own voice, but that feature will come in a later version of the product.) From there, the call functions as a regular Skype voice call.īut there’s a big difference. When you initiate a Skype Translator call, you select a male or female voice to work as vocal proxy.
“And this is machine learning, so the more usage we get, the better it gets.” “Our goal is to have every human in the world able to use this tool on whatever device they have,” says Gurdeep Pall, corporate vice president for Skype. In the very near-term it could change the way individuals and companies across the world interact. In developing such a platform, Microsoft Research (MSFT) has not only solved an extremely difficult computational problem that has for years dogged academics, researchers, and even DARPA, the research arm of the U.S.
(Microsoft acquired Skype in 2011.) The tool aims to be something like the Universal Translator from Star Trek: one language goes in and another language comes out, allowing two speakers who know nothing of each others’ tongues to interact in normal, if slightly stunted, conversation. Skype Translator is more than a decade of behind-the-scenes work at Microsoft Research.
0 kommentar(er)
