Harmony Central Forums
Announcement Announcement Module
Collapse
No announcement yet.

App for audio feedback frequency detection (giving the EQ band)?

Page Title Module
Move Remove Collapse









X
Conversation Detail Module
Collapse
  • Time
  • Show
Clear All
new posts

  • App for audio feedback frequency detection (giving the EQ band)?

    TL;DR: Seeking advice with building a visual audio feedback detection app for mobile devices.

    Hi, my name is Bohumir and I'm a software engineer and amateur musician. As a hobby project I'm building an application that detects and visualizes musical tones in real-time. I've realized that another app based on similar principles might be useful to sound engineers.

    So for curiosity I've started writing a mobile/tablet application that would detect when audio feedback occurs on the live stage and what's the precise feedback frequency(ies). This in turn might help the sound engineer (SE) manually turn down the corresponding band(s) on the EQ.

    And I'm wondering if anyone else is interested in this problem. I'm asking myself the following questions regarding the fight with feedback and I'd like to ask you for advice.

    1. Does the frequency need to be detected exactly or just selecting its corresponding EQ band is sufficient and more interesting?
    2. Is for less experienced SEs detecting the correct band by ear hard (neither fast, not precise) and more more experienced ones a lot easier?
    3. Does even for experienced SEs detection by ear get worse if they're tired after many hours of work and/or would they like to confirm his guess by an exact measurement?
    4. There exist some general purpose real-time spectrum analyzer hardware units and mobile apps. Do you thing a specialized app that would directly tell you the corresponding band on your EQ (in addition to the precise frequency) would be more beneficial?
    5. Also there exist some expensive hardware feedback eliminators that automatically adjust the EQ on your behalf. Many people tell that they are too aggressive in killing feedback at the expense of making the sound too dull. Do you thing a more lightweight and not fully automated thing like a mobile app could serve better?
    6. Do you thing that maintaining the list of feedback frequencies that occurred during the live performance (eg. after the initial setup) would be useful? Eg. to quickly eliminate any of those if it occurs again.
    7. Is it important that the feedback detection distinguish feedback squeaks from ordinary instrument and voice tones so that the music tones are not regarded as feedback?
    8. Is the almost real-time responsiveness (eg. delay of just several milliseconds) really important?
    9. Do you think that it is critical that the app just gets along with the built-in microphone or connecting to an auxiliary audio output from the mixing console is not a problem?
    10. Would a phone or a tablet with bigger screen be better suitable for such an app? Or it doesn't matter?

    I'd be very grateful if you could look answer or just discuss any one of those questions. Besides you raise your karma I could provide you with the application since the early development and fine tune the hand-crafted app to your specific needs.

    All in all, feedback detection seems to be quite an interesting problem to solve and I'd be glad to hear if such an app helps saving somebody's time or maybe even a live performance

    Thanks a lot!

    Bohumir


  • #2

    el_griton wrote:

    TL;DR: Seeking advice with building a visual audio feedback detection app for mobile devices.

    Hi, my name is Bohumir and I'm a software engineer and amateur musician. As a hobby project I'm building an application that detects and visualizes musical tones in real-time. I've realized that another app based on similar principles might be useful to sound engineers.

    So for curiosity I've started writing a mobile/tablet application that would detect when audio feedback occurs on the live stage and what's the precise feedback frequency(ies). This in turn might help the sound engineer (SE) manually turn down the corresponding band(s) on the EQ.

    And I'm wondering if anyone else is interested in this problem. I'm asking myself the following questions regarding the fight with feedback and I'd like to ask you for advice.

    I will answer within the quote block rather than waiding through the HTML formatting syntax here. Yes, it's interesting to the industry as a whole and there are many patents that I am aware of the deal with feedback detection algorithems, both hardware and software based. It's a very, very difficult task to generate your own algorithems, and even harder if you expect that they will work well. Do you hope to make this work or indicate dynamically while music is playing?

    1. Does the frequency need to be detected exactly or just selecting its corresponding EQ band is sufficient and more interesting?

    Probably best done as a function of both, weighting the indication that may be centered off of the eq's center frequency, so that a feedback frequency that is between 2 eq centers would indicate both bands are involved, and an indication proportional to the location relative to the 2 centers of adjacent bands would be helpful.

    2. Is for less experienced SEs detecting the correct band by ear hard (neither fast, not precise) and more more experienced ones a lot easier?

    Yes, I think this is accurate.

    3. Does even for experienced SEs detection by ear get worse if they're tired after many hours of work and/or would they like to confirm his guess by an exact measurement?

    Yes, though most (really) experienced guys have al the necessary tools and more importantly understand how to use them.

    4. There exist some general purpose real-time spectrum analyzer hardware units and mobile apps. Do you thing a specialized app that would directly tell you the corresponding band on your EQ (in addition to the precise frequency) would be more beneficial?

    Probably not, It's just another way of indicating the same thing

    5. Also there exist some expensive hardware feedback eliminators that automatically adjust the EQ on your behalf. Many people tell that they are too aggressive in killing feedback at the expense of making the sound too dull. Do you thing a more lightweight and not fully automated thing like a mobile app could serve better?

    Good question. Feedback eliminators require spohisticated algorithems to manage this. Some units do a better job than others, but the best results occur on already well tuned systems which most less experienced users do not have. This is one reason, IMO, for the poor results often experienced. They are also seen as a cure-all, which they are not.

    6. Do you thing that maintaining the list of feedback frequencies that occurred during the live performance (eg. after the initial setup) would be useful? Eg. to quickly eliminate any of those if it occurs again.

    Most experienced engineers probably do this as a 6th sense, automatically they have a good feeling where the problems are and are aware of this.

    7. Is it important that the feedback detection distinguish feedback squeaks from ordinary instrument and voice tones so that the music tones are not regarded as feedback?

    Absolutely, and I think this alone will be the "elephant in the room" regarding the amount of effort, math and coding experience required. Very, very difficult. I'm sure Don would agree here, as he has at least as much experience with this type of product as I do.

    8. Is the almost real-time responsiveness (eg. delay of just several milliseconds) really important?

    Real time FEEL is important, though real time for practical purposes may be on the order of a couple HUNDRES mSec. Several mSec would not be anywhere near enough time to detect and descriminate a feedback event because that would mean it would have to complete the cal;culations within 1/2 cycle at 6kHz which is not even enough time to gather the necessary data on which to perform calculations.

    9. Do you think that it is critical that the app just gets along with the built-in microphone or connecting to an auxiliary audio output from the mixing console is not a problem?

    Yes, being a stand alone tool is probably critical to it's success, especially for a non-experienced user.

    1. Would a phone or a tablet with bigger screen be better suitable for such an app? Or it doesn't matter?

    I prefer larger for ease of use, but the "kids" these days don't seem to have a problem making gross mistakes because what they thought they saw on their tiny device was not what was really there. Spent much of yesterday dealing with techno-geeks who were so enamored with their technology that they missed some really big things... didn't seem to matter much to them that they blew it in front of a crowd either, but in the professional world thet would have cost somebody their job.

    I'd be very grateful if you could look answer or just discuss any one of those questions. Besides you raise your karma I could provide you with the application since the early development and fine tune the hand-crafted app to your specific needs.

    All in all, feedback detection seems to be quite an interesting problem to solve and I'd be glad to hear if such an app helps saving somebody's time or maybe even a live performance

    It is a fascinating field of study, and a compete specialty unto itself. The more you learn, the more you realize that you don't know! Hope this helps.

    Thanks a lot!

    Bohumir


     

    -------------------------------------------------------------------------------------------------------------------------------------------------------------------
    Former product development engineer: Genz Benz, a KMC Music/Fender Musical Instruments Company, continuing factory level product support and service for Genz Benz

    Currently product development engineer: Mesa Boogie

    Comment


    • dboomer
      dboomer commented
      Editing a comment
      Lotta questions here

      Judging by them you have a long way to go to get to anything that will be useful. I don't mean to sound rude, just succinct. I would suggest you start by reading the tech paper on Rane's website and then looking up the Sabine patent ( which is now expired). These will give you better understanding of the mechanisms in play.

  • #3

    So for curiosity I've started writing a mobile/tablet application that would detect when audio feedback occurs on the live stage and what's the precise feedback frequency(ies). This in turn might help the sound engineer (SE) manually turn down the corresponding band(s) on the EQ.

    Precise is impossible to grab for many in the pro audio field.

    A parametric (PEQ) is about as precise as it gets for instant control.

    Most do not use parametrics that I have seen in 35years.

    Like Aged said...two sliders is about what most deal with for frequencies "in between" on a 1/3 octave GEQ.

    Thats a big task for a small window.

    Best of luck on that venture!

     

    Comment


    • dbMontana
      dbMontana commented
      Editing a comment
      Aged and Don would be the most knowledgeable on this (by far) and I see they've both responded but I assume you are aware that there a several good FFT apps available which, while not automatic, do a pretty good job of identifying frequencies that are "taking off" so to speak.

    • Dogoth
      Dogoth commented
      Editing a comment

      Pro Sound Guy wrote:

      A parametric (PEQ) is about as precise as it gets for instant control.

      Most do not use parametrics that I have seen in 35years.

      Like Aged said...two sliders is about what most deal with for frequencies "in between" on a 1/3 octave GEQ.

      Best of luck on that venture!

       


      I use parametric EQ all of the time. As a mater of fact If I have a decently tuned system, I prefer it as it can be exactly that "more precise". I currently use a 1/3 octave RTA and often wish I had the exact frequency (because a digital parametric notch is steeper and has less artifacts (phase junk) than an analog graphic EQ. With more & better digital filters coming along, I see more young engineers forsaking 1/3 octave GEQs for parametric. If you can't fix the issues you have with 4 bands (and many boards offer more on the output side), you need to step back and find what is wrong with your system/room/source and fix it (this can often include placment).

      I'm with agedhorse on this one in that I prefer a large/bright screen but that's just because my eyesight isn't what it used to be (glad my hearing is still adequate :-).

      Agreed - Best of luck with your design.

      Cheers


  • #4

    Big thanks, gentle people, for all the valuable replies, I appreciate it a lot! Also thanks for good wishes which give me some more energy to go on :manhappy:. Let me add some more information.

    The key point was to find out more about the problem and how you deal with it. However, I'd like to specify some things about what the application should be doing as it might not have been apparent. Think of it as a "smarter spectrograph". It should not neither replace the equalizer, nor automatically modify the EQ settings. The main differences to a plain spectrograph are the following:

    • Detecting frequencies of feedback (during its onset and course) among the signal
    • Discriminating them with normal musical tones (to prevent false positive alarms).
    • Detecting the frequencies precisely. Not just to the nearest 1/3 octave bin or FFT bin, but to about 0,0001-1 Hz. Also giving precise frequencies in the low-end (basses) - eg. up to 10 Hz. The purpose is to give precise data for setting parametric equalizer (if needed) and to obtain good input data for recognizing feedback among musical tones reliably.
    • Detecting multiple such frequencies at once.
    • Giving the sound engineer a (graphical) hint which band on a graphical equalizer to manually set based on the frequency. The particular equalizer type could be selected beforehand. The purpose is that the SE can reduce the reaction time of setting the EQ or just could use the tool to assure that his guess is alright.
    • Maintaining a overview of feedbacks that have been detected and suppressed during the whole show.

    I've looked at many applications out there that could possibly do the task, but none of them fulfills those requirements.

    Yes, I know the problem is not trivial and I'm not a domain expert in audio engineering. Certainly noone, including me, is able to grasp all the details of all the engineering and scientific areas involved. But on the other hand I have some background which could be beneficial to get this particular project done. About two years ago I graduated with an MSc. in Computer Science with specialization to Computer Graphics, Image Processing and Pattern Recognition. During the last several years my job has been in various areas of software engineering. For last about two years I've been studying various scientific papers regarding computational music processing as a hobby. In fact I've been asking questions about music processing as long ago as on high school and it was one of the reasons why I started studying Mathematics and Computer Science.

    For now in my hands I have a working prototype of an algorithm for very precise frequency detection (generally 0,00001 - 1 Hz) even for frequencies as low as 10 Hz (from a single frame). It follows several about 20-years-old papers. Also during long winter nights I've build a working software for real-time musical tone detection and visualization. The guts are very similar and it's expected that the feedback-related software would behave similarly. What is the interesting part is creating a new custom algorithm for recognizing which frequencies are candidates for feedback and discriminating them from ordinary musical tones. Currently, I have many ideas on the paper of how this can be accomplished.

    The only thing I'm trying to solve before delving deeper into those pattern recognition algorithms is whether the resulting application would be indeed useful and under what circumstances. That's the reason for my questions. I'd be grateful for hearing any further feedback (except audio feedback, tough :manwink:).

    Bohumir

    PS: If you'd like look at the attached images which show some examples of the precise frequency detection algorithm prototype.

    high-mid-freq-under-slight-noise.pngweak-mid-freq-under-very-heavy-noise.pngstrong-very-low-freq-under-noise.png

    Attached Files

    Comment


    • agedhorse
      agedhorse commented
      Editing a comment

      Detection within 1/12 octave is plenty accurate for just about any practical application I can think of.



Working...
X