Post Go back to editing

AEC as a sidetone echo canceller

Category: Software
Product Number: ADAU1467

Hello everyone,

In a very unconventional way, we would like to use an Echo Canceller to suppress audio sidetone from an analog audio intercom.

Under normal circumstances, the audio intercom system compatible to wired headset generates a sidetone useful for the operator to hear his voice.

We are looking for a solution to reduce or cancel the sidetone thanks to a Line or Acoustic Echo Canceller.

The latency of the sidetone is low (I would say 5ms or less)

Do you think the AEC algorithm available in Sigma Studio is capable for this kind of job ?

Thank you 

Sebastien

Parents
  • Hello Sebastien,

    Is the sidetone generated in the actual headset? 

    I will assume the DSP is in the intercom belt pack or desktop unit correct? 

    I assume you have access to the near side and far side audio correct? 

    If the sidetone is actually generated in the headset it will be difficult to cancel out. You will not have the ability to generate an error signal to have a filter learn the signal.

    The acoustic echo canceller has to be able to generate an error signal. So I do not think this is a solution if the sidetone is generated in the headset. 

    Since the delay of the sidetone should be somewhat predictable and fixed. Then I think you can take the nearside audio, reverse the polarity, add some delay(if needed) and then insert it into the farside signal going to the headphone. Then the headphone will add the sidetone and the idea is that the signal you are adding will cancel it. 

    Dave T

Reply
  • Hello Sebastien,

    Is the sidetone generated in the actual headset? 

    I will assume the DSP is in the intercom belt pack or desktop unit correct? 

    I assume you have access to the near side and far side audio correct? 

    If the sidetone is actually generated in the headset it will be difficult to cancel out. You will not have the ability to generate an error signal to have a filter learn the signal.

    The acoustic echo canceller has to be able to generate an error signal. So I do not think this is a solution if the sidetone is generated in the headset. 

    Since the delay of the sidetone should be somewhat predictable and fixed. Then I think you can take the nearside audio, reverse the polarity, add some delay(if needed) and then insert it into the farside signal going to the headphone. Then the headphone will add the sidetone and the idea is that the signal you are adding will cancel it. 

    Dave T

Children
  • Hello Dave

    Thank you for your support Pray 

    The DSP will be installed inside an audio base station.

    This base station is connected by wire to the intercom (analog interface)

    A belt-pack is wirelessly connected to this base station

    A headset is connected to the belt-pack 

    The sidetone is generated by us in the belt-pack but another sidetone is provided by the intercom (it's not possible to disable it).

    Today with a LEC implemented in the base station we are able to suppress the sidetone from the intercom. But for some reasons, we must find another solution.

    The idea is to suppress in the the base station the audio loopback generated by the intercom.

  • Hello Eate77,

    Hmm... this is a bit confusing but I think I got it. Block diagrams really help to understand signal flow. I sketched this up:

    This is my interpretation of what you described. The part in RED is what I think could be done in the DSP to cancel out the sidetone. 

    You should be able to determine the delay either with a scope or by experimentation. You will also have to match the level, I did not show a level control on the cancellation signal. I would hope this timing would not change but this depends on what the intercom is doing. I would not expect it to change.

    Dave T

  • Hello Dave,

    our setup is exactly what you drew (appreciate your commitment)

    No the delay is fixed, as it's a sidetone, the delay shouldn't be long, otherwise, the operator will not be able to speak properly (as he listen his voice only through the headset)

    i'm going to test the trial mode with the example project.

    Regards,

    Sebastien