Nine steps to dramatically reduce unsteady noise in mobile phones

Figure 1. Time domain waveforms before and after using instantaneous unsteady noise suppression techniques.

1. Understand the difference between steady state and unsteady noise

Due to its relatively stable nature - such as in a noisy blower environment, steady-state noise can be easily identified and easily removed using traditional new good handling methods. In contrast, unsteady noise has a rapid or random change, such as a person's voice, background music, or touch tone. When unsteady noise is recognized as noise, these noises have actually passed, so a more sophisticated noise suppression method is needed.

2. Use two microphones to improve your understanding of the auditory scene

Next-generation noise suppression techniques such as Auditory Field Analysis (ASA), Beamforming (BF), and Blind Source Isolation (BSS) use several microphones to more accurately identify, locate, and group noise sources with higher accuracy than a single microphone. Today's handset manufacturers are aware of this trend, introducing a second microphone into the phone's architecture.

3. Use grouping principle to separate sound points

The grouping method simplifies noise suppression while also determining the source of unsteady noise. For example, Auditory Field Analysis (ASA) uses the human auditory channel as a model to process noise according to the actual human hearing of a certain sound. By grouping multiple acoustic energy to reproduce the original sound, the ASA achieves accurate grouping from multiple sources while avoiding any sounds that should be separated by hearing. The grouping principle can be roughly described as continuous (occurring over a period of time) and simultaneous (within a range of frequencies).

4. Use multiple threads to group, otherwise it is difficult to group sounds correctly

Every grouping clue has limitations. Use multiple threads to properly group sounds that are difficult to analyze. Some important clues include:

• Characteristics of sound: Harmonics generated by a source can form a clear frequency map that can be used to distinguish between two sounds. Sound characteristics are a major clue to distinguish between men and women.

• Spatial information: The distances determined by distance and direction can be used to group sounds so that the sounds of interest can be distinguished.

• Time of issue: If the two sound energies and their corresponding harmonics are consistent in time, they are likely to come from the same source.

5. Reduce convergence time to eliminate more transient noise

Traditional noise suppression methods must converge before they cancel noise, so they are less efficient at suppressing unsteady noise sources. By using a fast-responding clue to describe the sound, instant moments such as snapping fingers can be identified and eliminated.

6. Use log-to-linear frequency ratio (FCT vs. FFT)

The familiar Fast Fourier Transform (FFT) decomposes the frequency components in a linear ratio, which limits the spectral resolution at low frequencies and uses a very constant frame size and frequency-independent bandwidth. In contrast, fast cochlear transformation (FCT) is based on the principle of the human cochlea and varies in logarithmic frequency. As such, it does not limit the spectral resolution. Through continuous processing rather than frame processing, FCT also reduces processing delays, so it is suitable for identifying unsteady noise sources. In addition, the bandwidth of the FCT operation is frequency dependent and can more accurately match the time domain-frequency domain tradeoff within the human ear hearing range.

7. Use omnidirectional microphones to reduce costs

Methods like beamforming require a specialized cardioid microphone. The cost of this type of microphone is higher than that of omnidirectional microphones, with lower noise margin, and must be individually calibrated and matched with an error of less than 1 dB, which limits the spacing and is sensitive to wind and breath. Therefore, it will increase the noise by +12dB. Beamforming techniques also have their limitations, and any erroneous choices of the beam of interest are erroneously transmitted. The management of the number of microphones required for a system is also important. For example, Blind Source Separation uses a simple linear non-mixing method that works well with as many microphones as there are sound sources.

8. Treat echo as an independent source

Echo cancellation methods have been used to eliminate echoes. This method is very computationally intensive because the echo reflection must be calculated and its performance is poor when the noise source changes quickly. Grouping clues allow us to use waves as another source of noise. Since the echo does not need to be calculated and does not change the path, it can perform instantaneous suppression and achieve echo noise suppression performance of up to 46 dB.

9. Adopt new test standards

The mobile communications industry is continually advancing test standards to demonstrate higher levels of speech quality through innovation in noise suppression technology. To ensure the best quality of their products, the industry recently revised the ITU P.835 specification to provide a consistent test methodology, testing voice quality with noise suppression techniques and generating test reports.