Receive Audio Processing

This page describes the Receive Audio Processing built-in to wfview

wfview receive audio processor showing Audacity noise reduction

wfview has sophisticated noise reduction and audio equalization which is available to users of Network (Ethernet or WiFi) radios. We plan to eventually introduce these features for USB-connected radios, but at this time, it is only available for Network connections. This includes connections made to remote wfview server instances with USB radios.

To access the Receive Audio Processing controls, press the “RX Audio Proc” button at the bottom of the main wfview window. This button is only functional when you are connected to a radio with an audio stream. It’s recommended that you choose a “1ch” audio connection format. When you choose a 2 channel format (such as on the IC-9700 and IC-7610), the audio processor will let you select which channel to process at the top.

The RX Audio Processor chain looks like this:

Radio Audio –> Noise Reduction (Speex or ANR) –> Equalizer –> Output Gain

Selection and adjustment of the noise reduction algorithm is critical to success. Careful adjustment will assist in the reception of weak signals and will also greatly enhance strong signals. Poor adjustments can make things sound worse.

Here is a basic description of each type of Noise Reduction offered.

ANR: Audacity Noise Reduction

The Audacity Noise Reduction algorithm is a modified version of the noise reduction code within the Audacity audio editor. The original Audacity code was written by Dominic Mazzoni and subsequently rewritten by Paul Licameli. The algorithm requires that a “sample” of the noise be taken. You can do this by first tuning off to an empty space on the band, and then pressing “Collect Noise Sample”. This will collect a sample for five seconds (you can stop it earlier if you like). The sample will be visualized in the noise profile plot and immediately activated. wfview will also save the sample for this mode so that you don’t have to take a fresh sample next time — but note that superior results will occur if new samples are taken as band conditions and receiver adjustments can alter the noise profile.

Once the noise sample is taken (and we need one for each receiver demodulation mode), the noise profile will be used to determine if a given spectral band likely contains speaking or noise.

ANR Adjustment strategy:

Set the Noise reduction slider to reduce by 30dB initially. We will set it to less in just a moment.

Set the Sensitivity slider to the far-left position. The audio will sound a bit like wind chimes. Now advance the Sensitivity slider towards the right. Observe that suddenly the “wind chime” sound goes away and the audio becomes quiet. Move the slider just a touch to the left to allow some of the artifacts in but not too many.

Now adjust the Noise Reduction slider to a comfortable value between 10 and 20 dB. This is how much the algorithm will reduce the noise. Most people prefer to hear a little of the original audio.

If you find that strong voices are being reduced, adjust the Sensitivity control a bit to allow for more artifacts but also for more sensitivity to voice.

Speex: Ephraim and Malah Minimum Mean-Square Error (MMSE)

The Speex MMSE noise reduction algorithm works by dividing the audio into a number of “Bark bands” (Bark refers to the actual spectral bands of the ear). Each band has the variance (energy of randomness) measured. Highly random bands are deemed more likely to be noise; lower entropy is more likely to be voice or other signals. Speex was written by Jean-Marc Valin, and our miniature noise-reduction-only implementation is available here on gitlab.

Speex Adjustment Strategy:

Set the suppression at 10 to 20dB.

Set the bands at 24 bands. More bands rejects more noise but also causes more artifacts. Cochlear implants use 24 bands, and that seems to strike a good balance.

Don’t enable AGC, this will actually pump the noise up between words and is somewhat self-defeating.

Set the following controls for typical SSB reception:

  • Enable VAD. This helps the algorithm understand syllables
    • Prob-Start: 85%
    • Prob-cont: 65%
  • Adaptation Speed:
    • SNR decay: 0.08 or lower
    • Noise Update: 0.01
    • Prior base: 0.26

For Short Wave Listening, you may want to make the timing less aggressive (slower).

Gain Adjustment

Once the Noise Reduction algorithm is active, you will see that the Input and Output traces on the spectrum are no-longer aligned. The Output should be lower than the input when there is noise. This is key. When a person speaks, the vocal spectra should rise above the orange and green lines. Adjust the Output Gain control until the background noise is below the (unprocessed) input level, and vocal peaks are visibly above the (unprocessed) input level. This condition raises the SNR, which is the entire point.

Equalizer

The equalizer provides four bands of adjustment. It is based on the Triple Para EQ from Steve Harris. The first and last band are known as “shelving” and are similar to traditional bass and treble controls. These controls raise or lower the signal level at their prescribed frequencies but also at frequencies which stretch beyond their setting. For the left-most control, frequencies at and below the prescribed frequency are effected. For the right-most control, frequencies at and above the prescribed frequency are effected.

For the middle two bands, only a portion of frequencies in the neighborhood of the selected frequency are adjusted. This is the “Q” of the filter, which is not adjustable but is preset at 0.8.

For those with hearing loss, you may find it handy to boost frequencies around 1-2 KHz by a few dB. Be mindful that you do not “go into the red” and exceed the digital audio limit. You can watch the main RxAudio meter in wfview or look at the spectrum. Or just listen for that nasty digital crunchy sound.

If you have modern “PC speakers” on your setup, you may want to roll off a few dB starting at around 200 Hz with the bass shelving control.

 

wfview receive audio processor showing Audacity noise reduction wfview receive audio processor showing speex noise reduction