Comparison with other candidate
technologies for noise cancellation
1. Audio
recordings of speech in high-noise
We present a comparison of sound recordings made in high noise (Blackhawk helicopter noise, about 94 db SPL) between our microphone housed within the NCA (theBoom) and an active noise-cancelling microphone (the jawbone). An utterance was recorded first through the jawbone and then through theBoom while noise recorded from a Blackhawk helicopter was played in the background at 94 db SPL. Upon listening to the recorded audio from both microphones, we find that:
- In the recording made through theBoom, the speech is clearly intelligible over the noise and the noise is barely perceptible in the presence of speech.
- In the recording made though the jawbone, the speech is intelligible, but sounds distorted. The noise is lower than the speech but is perceptible even in the presence of speech.
Figure 1: Waveform of speech recorded in Blackhawk helicopter noise though the jawbone microphone
The jawbone signal’s waveform (Figure 1 above) shows some clipped spikes that distort some of the features of the spectrum, which are more apparent in the spectrogram for the same audio recording. The speech signal is clearly higher than the noise, which indicates that the microphone does have noise cancelling properties.
TheBoom’s signal waveform (Figure 2 below) does not exhibit clipping. Also, we can see that the speech signal is much higher than the constant background noise compared to Figure 26. This indicates a higher S/N in theBoom’s recording.
Figure 2: Waveform of speech recorded in Blackhawk helicopter noise though the Boom microphone
3. Comparison
of Spectrograms
Figure 3: Spectrogram of speech recorded in Blackhawk helicopter noise though the jawbone microphone
Figure 4: Spectrogram of speech recorded in Blackhawk helicopter noise though the Boom microphone
The jawbone signal shows some significant losses in key areas of the spectrum. This can be clearly seen from the bright yellow bands below 4 kHz which correspond to the vowel sounds. These appear brighter and clearer in Figure 4 (theBoom). We also see bright yellow parts more distributed in Figure 3, whereas they are more concentrated in local regions in Figure 4. This indicates clipping in the signal recorded through the jawbone, which typically distorts critical features within the spectrum for speech, especially for vowel sounds. The noise component is eliminated to a greater extent in theBoom. Note also that noise (represented by the loosely distributed areas of red and purple rather than the isolated bands of red), especially above 4 kHz is much more evident in Figure 3 than in Figure 4, indicating the more dramatic noise cancelling properties of the NCA on theBoom over the Jawbone.
TheBoom’s noise cancelling properties are excellent and do not negatively contribute to overall speech clarity. The spectral properties features of speech are preserved with theBoom under conditions of significant noise, while the same features as recorded through the Jawbone are significantly blended under conditions of significant noise, thereby negatively affecting target signal clarity.
The near-field and far-field signals have a lot of overlapping frequencies. This makes it extremely difficult to actively remove the noise without removing important speech frequencies. Some designs (like the jawbone) attempt to enhance active noise cancellation by using technologies to identify vocal patterns and isolating them. However, due to the inherent frequency distorting nature of active noise cancellation, the speech signal is invariably altered in an undesirable way from the point of view of automatic speech recognition (ASR).
TheBoom uses acoustic passive noise-cancelling technology that cancels out noise before it enters the signal. This is inherently better than attempting to remove the noise from the signal after it is already mixed in. A high S/N is attained without sacrificing the flat frequency response.