NEC Corp developed a technology to reduce wind noise and tapping sound with only one microphone as part of its "EuphoMagic" acoustical signal processing technology.
The company demonstrated the technology in the booth of the NEC Group at the 16th Embedded System Expo (ESEC), which runs from May 8 to 10, 2013, in Tokyo.
The technology enables phone calls and recordings with natural voice while reducing wind noise and tapping sound when a mobile phone, smartphone or camcorder is used outdoors. In the market, there are some devices that have a function to reduce such noise, but they remove all the frequencies of noise, reducing voice volume, too, NEC said.
On the other hand, the new technology reduces only noise, making it easier to hear voice and record it clearly. At the booth, NEC demonstrated the technology by using a smartphone. And it was possible to check the effects of the technology by switching on/off the noise reduction function.
Because the technology can be used for a smartphone, whose computing power is limited, it can be applied to a variety of devices, the company said.
Wind noise decomposed into stationary, nonstationary noises
The details of the new technology reducing wind noise are as follows. It considers that wind noise consists of (1) a constant noise generated by wind (stationary noise) and (2) a noise changing depending on the strength of wind (nonstationary noise) and appropriately reduces them.
This is the world's first technology that reduces only wind noise by flexibly responding to the characteristics and strength of noise, NEC said.
Stationary noise is calculated from the average value at the time when input signals (sound coming into the microphone) do not contain voice. And nonstationary noise is calculated from the difference between the input signals and the estimated value of stationary noise. These estimated values are removed from the input signals.
When voice volume abruptly increases, it tends to be mistakenly detected as nonstationary noise. But the new technology correctly recognizes such a voice based on the sizes of the waveforms of its frequencies. Specifically, the variation in waveform size at several different frequencies is examined. And the technology considers sound with a large variation as voice and sound with a small variation as nonstationary noise.
Furthermore, to reduce the amount of calculation, the process of decomposing and synthesizing input signals into several different frequencies is shared for the reductions of ambient noise and wind noise. As a result, the amount of calculation was reduced by about 30%, compared with the case where the reduction processes are conducted separately. So, the technology can be used for devices with a limited computing power such as smartphones without delay, NEC said.