I'm working on developing an audio device testing suite (for speakers and like). I'm looking for approaches to detecting two types of noise: static such as white or pink noise and artifacts such as crackling, 'wheezing' etc. I'm kind of new to the DSP field and was wondering how I should approach this task? The plan is to use a measurement microphone and send a known signal (such as a sine wave or a sine sweep) through the speaker, and then analyze the recording.

My immediate thought process is to play a sine wave or even a sine sweep and look for undesirable frequencies via a spectrogram (some noise levels should be acceptable), but I feel this might be overtly simplistic and I might be missing something.

I was not able to find many resources on this topic, so any help or pointers would be appreciated!


1 Answer 1


That's a tough problem to solve. There are commercial solutions for this type of measurements (e.g. Audio Precision (ap.com), Listen, Inc. (https://www.listeninc.com/). They do an ok job but tend to be expensive since they had to invest many thousands engineering hours to develop the algorithms and measurement systems.

There are sizable number of artifacts that can be detected and each typically requires its own highly specialized measurement

  1. Noise floor. Typically done with calibrated microphone, with and without broadband excitation
  2. Non-linear distortions: can be done with sweep signals in a well controlled environment by tracking fundamentals and harmonics
  3. THD (Total Harmonic Distortion and Noise). One frequency at a time, gating out the fundamental
  4. Rub and Buzz: done with sine sweeps. Still mainly done by human listeners but there are some newer algorithms including machine learning that are doing allright
  5. Airleaks: Low frequency sine wave excitation with human listener (often using a stethoscope). Can also be determine by pressurizing the enclosure and measuring the time constant of the pressure equalization
  6. Thermal compression: broad band excitation at different levels and excitation times.
  7. Crackling, skipped samples, spurious clipping: I'm not aware of any systematic measurement: It's typically detected by ear and then analyzed looking at the wave form.

It turns out that human perception of these artefacts is really complex and humans are quite good at it. That's why most production facilities (where time is money) still use human listener for production line testing, despite the considerable cost in terms of labor, time, and facilities.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.