Apple has Siri, Google has Google Now, and Microsoft has a promise. While we wait for a true personal assistant on Windows Phone, we can look forward to this leaked video of how voice searches might work on Bing in the near future.
Microsoft's Tellme technology has been integrated with Windows Phone since the beginning. It’s gotten better over time too, slowly adding new features like text-to-speech for text messages. A ‘leaked’ video from TechFest earlier shows off some new features and improvements to come for the speech recognition technology. The video below compares two Windows Phone running a custom Bing app. The one on the left targets the old speech recognition backend, while the one on the right reveals the newer system.
- New system is faster. Time difference of 1.01 seconds vs .56 seconds for when he stopped talking to when the results came back.
- Faster results enable results to stream as you speak
- Accuracy has improved when noise is introduced to the environment
An interesting note a few sites have glossed over. In the video above, the phone on the right is using the newer backend system for speech recognition but it’s running on older data center hardware. Recently they’ve updated the hardware for the current speech recognition backend. If the old backend was running on the same hardware as the newer system, the same search would actually take 1.51 seconds. So if you’ve been noticing faster results when using Bing voice search, now you know why. When we they finally put the newer backend in main use on the new hardware, results should be slightly faster than in the video.
Microsoft Research has been the driving force behind these improvements, especially in the area of improving accuracy. By basing the speech recognition off of deep-neural-network (DNN) algorithms, they’re able to achieve a 10-20 percent improvement in error reduction and use 30 percent less processing time. DNN also allows these improvements without needing the user to adapt. The long term goal is to make a product that allows for out of the box speech recognition without forcing the user to adapt to the product. If you want to read more about Microsoft Research and their deep-neural-network technology, go here.
We don’t have any timetable for when we can start seeing the newer Bing voice search; your guess is as good as ours. But if I was to make a wager, I'd probably bet on its inclusion in the next big update – maybe Blue.