Unlocking the secrets of NEC's "EuphoMagic" acoustic signal processing technologies, which make conversation easy and dramatically improve voice recognition rates even in noisy environments
Communicating audio to machines
- MitaNow I have some questions for you, Mr. Tsujikawa. Can you tell us about the voice recognition used to communicate audio to machines?
- TsujikawaFirst, allow me to give a brief overview of how voice recognition works. Machines for recognizing voices have already learned the frequency ranges in which individual phonemes are found. This process involves collecting and learning thousands or tens of thousands of pieces of sample data. The machine references this data and estimates which the voice audio input matches.
- MitaSo that's how it's done.
- TsujikawaThe removal of noise also ultimately involves the machine adjusting voice audio to bring it closer to the form that it has learned in its database. That means there is a difference between voices that are easy for a machine to recognize, and voices that are easy for a human to hear.
- MitaI see.
- TsujikawaIn the same way, our noise suppression technology for voice recognition shapes and processes the signal from the microphone to make it easier for machines to recognize voices. NEC products that incorporate voice recognition are suited to a variety of applications, including converting audio at meetings into text data in real time*1, and enabling voice input for tasks such as inventory control at factories*2. NEC's voice recognition technology has also been used in a service at Sanrio Puroland (Tama, Tokyo) that lets you to talk with a Cinnamoroll character*3. Proper shaping and processing for voice recognition is what enables products and services in noise-filled environments like factories and amusement parks.
- *1 A product called "VoiceGraphy"
- *2 A product called "VoiceDo"
- *3 The "Oshaberi Cinnamon" service that ran from April to June in 2012
- MitaYou know, I often use voice recognition on my smartphone, but when there is noise in the background it tends to get things wrong a lot of the time. Is there anything that can be done to make voice recognition work better in noisy places like this?
- TsujikawaWe introduced that technology in May of this year. In the past it was possible to perform voice recognition with the microphone close by in some cases, but the technology we developed recognizes voices even when speaking a short distance away from the microphone. This technology removes noise using dual microphones, and then optimizes the audio for voice recognition. This enables voice recognition to be performed using smartphones and tablets even in noisy places. It is also possible to perform voice recognition in locations with background noise, such as moving vehicles or living rooms with the TV on.
- MitaCan that really be done?
- TsujikawaWhy don't you see for yourself? (He gets out acoustic devices such as speakers and a PC). I'm about to play the sound of a crowd through the speakers. There's a microphone here, so can you try to say something from a short distance away, Ms. Mita?
- MitaOh, okay. "Hello. It's a pleasure meeting you." (The device repeats what she said).
- MitaWow! It spoke back to me!
- TsujikawaThe device recognized your voice among the background noise, and repeated it back to you. This shows how audio spoken away from the microphone in a noisy environment is still recognized correctly.
- MitaThis technology is just what I was looking for! I hope to see it used in all sorts of devices. Thank you for filling us in on this interesting topic today!
- Miyahara, TsujikawaWe're glad we got a chance to speak with you. Thank you for hearing us out.
In this installment of the program we heard about EuphoMagic, which makes phone conversations and voice recognition easier in noisy places. I got to see for myself how noise-free and clear the audio became. I only wish I could fully convey the difference to you all. I'd like this technology to become more widely used, so it makes our lives easier! See you in the next installment of "MiTA TV"!