You happen to have a 250 kS/s multichannel DAQ lying around? What's the resolution? A "96K recorder" is nowhere close to sufficient to do phase and transient measurements on audio signals. I'd probably prefer something closer to a 1 MS/s sampler.

I suspect that if you think the software will take anything less than a couple of weeks to write, you haven't thought about the actual analysis required. I based my estimate on the six weeks that I would quote a client to put that system together; I could do it for my own curiosity for significantly cheaper, but it would be neither professional nor conclusive.

Quote:



We're talking about burn-in testing, which should be a measurement of audible differences in sound. There is nothing audible about sound that can't be picked up with a 96 Khz sample rate. It gives enough information in both the frequency and time domains to do phase and transient measurements up to the limits of the human ear. Why do you think it doesn't?

Also, you'd need something stable enough to record hundreds of hours of continuous audio. Standalone recorders are that stable. Computers generally aren't. My approach would be to record the audio signal to a standalone recorder, and then transfer the files to a computer over ethernet.

I agree that the software would probably be on the order of a month full time or a couple of months part time. But you may be able to cut that down pretty significantly by finding an audio analysis program that has an API, or leveraging open source libraries. Licensing for release wouldn't be a concern if the software was only for internal use.

Barg. Head-fi keeps eating my response.

I'll try one more time. You want to oversample for several reasons: 1) noise reduction, so that you can utilize the full dynamic range of the mic and ADC; 2) there are some studies that indicate that the brain is sensitive to timing that is significantly faster than the limits of the ear, this is supposedly data that is used in creating spatial awareness for sounds; 3) phase and transient analysis would be much easier with 5-10x oversampling than with 2x.

As to stability, there is no reason that a computer based test system can't be used. I've implemented many test systems with hundreds or even thousands of hours of continuous active testing. Also, when dealing with such a contentious topic, it's best to remove all possible sources of skepticism from the test protocol. Using an automated test system makes it much easier to verify that every UUT was tested in exactly the same fashion.

Last, I still think you're underestimating the analysis a little bit. It's not sufficient to just compare the waveform output from time A to that from time B. Yes, if they're exactly the same, you've disproven burn-in. However, if there are differences, you need to be able to analyze them and characterize the changes. A good test protocol should probably include several different test components, utilizing pure frequencies, sweeps, and then complex sounds, and each of those sections should be analyzed individually (and in different ways), then an amalgam report generated. It could be less than a month, but if I were quoting this project fixed-price, I wouldn't bet my wages on it.