intro algorithm sounds

Adaptive beamforming - sound examples

The following sound examples illustrate the speech enhancement capabilities of the SP-SDW-MWF adaptive beamforming algorithm in different noise scenarios. For each scenario two sound files are presented :

  • the unprocessed sound file, which is the signal captured by the front microphone of the right hearing aid. The front, middle and rear microphone signals of the right hearing aid are the inputs to the SP-SDW-MWF algorithm.
  • the processed sound file, which is the output of the SP-SDW-MWF algorithm.

Low-reverberant room with stationary directional background noise

  • low-reverberant room
  • female speaker in front of the listener (at 0°)
  • directional stationary noise source having the average spectrum of human speech at the right-hand side of the listener (at 60°)
  • average speaker signal level 5 dB above the noise signal level in the unprocessed condition (SNR=5dB)

unprocessed.mp3
processed.mp3

As can be observed, beamforming algorithms suppress directional noise sources very well in a lowly reverberating room.

Living room with a directional music signal as background noise

  • living room
  • male speaker in front of the listener (at 0°)
  • directional music source at the right-hand side of the listener (at 90°)
  • average speaker signal level equal to the noise signal level in the unprocessed condition (SNR=0dB)

unprocessed.mp3
processed.mp3

Despite the more difficult scenario due to a lower SNR and more room reverberation the amount of noise suppression is still considerable. Observe that the quality of the desired speech signal is hardly affected by the processing.

Cafeteria with babble noise in the background

  • cafeteria room
  • female speaker in front of the listener (at 0°)
  • diffuse cocktail-party noise (crowded cafeteria)
  • average speaker signal level equal to the noise signal level in the unprocessed condition (SNR=0dB)

unprocessed.mp3
processed.mp3

This is the most challenging scenario : a highly reverberating room and diffuse background noise. Note that the amount of noise reduction is now clearly lower than in the first, low-reverberant scenario. Nevertheless, the algorithm is still capable of enhancing the signal without too much affecting the signal quality.