Audience Announces S1.0 for Windows x86
by Joshua Ho on November 5, 2014 1:34 PM EST
While smartphones have always had a strong need for quality voice processing, on the PC side such concerns have generally fallen to the wayside as voice was generally ignored, with only a few niche applications. However, as tablets and other formfactors have become more common on Windows x86, there is increasing demand for voice control and other features for a similar experience to Android and iOS tablets. To enable this, in conjunction with the eS800 Audience is announcing their S1.0 voice software, which delivers eS700-series voice and audio algorithms over Windows' audio interface.
There are a few key features that this enables, namely wideband noise suppression for VOIP, omnidirectional noise cancellation, key-click removal, and improved speech recognition. In the case of wideband noise suppression, this means that rather than artificially compressing the range of frequencies received to be able to cancel out noise, the software suite makes it possible to process incoming audio over almost the entire range of hearing for higher quality VOIP.
While omnidirectional noise cancellation seems to be a bit uninteresting, the real story here is that it's possible to enable a good voice experience without beamforming, which relies on two microphones that are aligned with each other. While this allows for effective noise cancellation of anything not collinear with the microphones, this makes it hard to avoid losing voice in multiple directions. As these algorithms enable noise cancellation without beamforming, this means that there's no need to drill holes in glass for microphone holes, which can be quite expensive as the holes cause yields on the cover lens to drop dramatically. This means that microphone placement in general can be more flexible, and only a pair of microphones is needed.
Although the eS800 line has more advanced voice processing features, the S1.0 suite still utilizes some level of artificial neural network technology to enable features like key-click removal. While details are a bit sparse on how speech recognition assist works, Audience emphasized that their solution had greatly improved recognition accuracy when compared to competitors. The S1.0 software will work on Windows 7 and 8.1 with Intel's Haswell and newer CPUs.
Facebook
Twitter
RSS
8 Comments
View All Comments
hansmuff - Wednesday, November 5, 2014 - link
"Key-click removal" would be so awesome to have when playing games, instead of having to use PTT. But how this would effectively filter out the various, different mechanical clicks is a mystery to me.A blue cherry switch sounds very different from an ALPS or an old IBM tactile. Perhaps this feature isn't really addressing this particular, small niche?
Murloc - Wednesday, November 5, 2014 - link
it would probably filter out any short impulse.Or maybe it would listen for impulses and if it detects a certain sound happening frequently, it will filter it out.
Or maybe it will have a bunch of settings so you move the cursors around until it filters out your specific keyboard.
Or maybe it has a training program which asks you to write with your keyboard for 10 seconds.
There's plenty of solutions for that and they've surely thought about the different click sounds.
epobirs - Friday, November 7, 2014 - link
A training mode can allow for sampling the distinctive noises produced by your system. It's still a necessity for the best possible voice recognition but nearly as much of a hassle as it once was.3DoubleD - Wednesday, November 5, 2014 - link
How is 80% word accuracy at library-level noise conditions even remotely acceptable? One in 5 words are unintelligible? That explains why I hate talking on cell phones/video calls/anything-that-is-not-in-person so much.Voice quality, whether it is on a cell phone, VoIP call, video conference, or during gaming needs some very serious attention. If an Android phone could boast a significant increase in call quality (both input and output), I'd seriously considering buying that phone. Call quality is quickly becoming my #1 concern.
Murloc - Wednesday, November 5, 2014 - link
I think it's about speech-to-text or voice assistants like cortana hearing you.hpglow - Wednesday, November 5, 2014 - link
The level of English modern Americans comprehend is just terrible. You can talk face to face with many people these days and have them not understand a thing. It's like they feed kids mercury and whiskey for lunch these days. Everyone can't be a genius but I really think something is making people dumber by the day.Alexvrb - Wednesday, November 5, 2014 - link
They're more interested in brainwashing children than educating them. You don't need to know proper English (reading and writing? wat a noob), math is for nerds, history isn't important... but you damned well better be a good little progressive. Oh and we're cutting down on your salt and ketchup intake. It's bad for you.dananski - Thursday, November 6, 2014 - link
Given it's words rather than % of words, I'd guess that it was an 80 word sample.Alternatively, maybe they had to talk really, really quietly since they were in a library.