Can You Hear the Difference? A Quick Blind Audio Test for a New Digital Audio System

[USF-Audio-2026]

Hello ASR members,

We are a group of scientists at Imperial College London working on a new digital audio sensing technology.
We have only recently been able to realize the full hardware and signal-processing pipeline.

As an initial proof of concept, we are running a short blind listening test that takes about 3–5 minutes, and we would be very grateful for your participation.

Audio Test: https://bit.ly/USF-Audio



Your feedback would be very valuable in helping us evaluate the perceived quality of the audio signals and guide our future research.

Many thanks in advance.

 

[Alan O]

Bit of a visual giveaway in one sample in each case, isn't there?

 

[Jarl]

Are the volume levels the same on each clip?
What is the purpose of a slider with so many values? What is score 51 compared to 52? I kinda wish you only had to rank them relative to each other.

 

[SSS]

Hello ASR members,

We are a group of scientists at Imperial College London working on a new digital audio sensing technology.
We have only recently been able to realize the full hardware and signal-processing pipeline.

As an initial proof of concept, we are running a short blind listening test that takes about 3–5 minutes, and we would be very grateful for your participation.

Audio Test: https://bit.ly/USF-Audio

View attachment 521184

Your feedback would be very valuable in helping us evaluate the perceived quality of the audio signals and guide our future research.

Many thanks in advance.

Made the test. One of the samples is alway absolutely distorted. The two other almost similar so it is not alway easy to judge the better one because also in these there are also slight distortions at least at my audio set.

 

[SSS]

Hello ASR members,

We are a group of scientists at Imperial College London working on a new digital audio sensing technology.
We have only recently been able to realize the full hardware and signal-processing pipeline.

As an initial proof of concept, we are running a short blind listening test that takes about 3–5 minutes, and we would be very grateful for your participation.

Audio Test: https://bit.ly/USF-Audio

View attachment 521184

Your feedback would be very valuable in helping us evaluate the perceived quality of the audio signals and guide our future research.

Many thanks in advance.

What means "digital audio sensing" in your case? Is it a microphone or acceleration sensor?

 

[USF-Audio-2026]

Bit of a visual giveaway in one sample in each case, isn't there?

Depending on the browser, yes. However, this will not impact the overall test which evaluates two different methods.

 

[SSS]

Would be nice if you may answer my question from my last post.

 

[USF-Audio-2026]

Made the test. One of the samples is alway absolutely distorted. The two other almost similar so it is not alway easy to judge the better one because also in these there are also slight distortions at least at my audio set.

Indeed that's the setup.

What means "digital audio sensing" in your case? Is it a microphone or acceleration sensor?

Sorry to miss this. It means a new method to digitize information.
The most classic method is PCM. However, one can devise their own method to digitize signals via new forms of ADCs.

 

[USF-Audio-2026]

Are the volume levels the same on each clip?
What is the purpose of a slider with so many values? What is score 51 compared to 52? I kinda wish you only had to rank them relative to each other.

The clips are played under the same setup, but they may not be perceived as equally loud, especially because distortion can affect perceived loudness.

The slider is not meant to make 51 vs 52 deeply meaningful on its own; it just gives listeners room to express finer differences. In the analysis, we look at trends across many ratings rather than individual one-point gaps. Ranking is a sensible alternative, but the current scale also captures how large the perceived differences are.

 

[SSS]

Thank you. Will the result be published here?

 

[USF-Audio-2026]

Thank you. Will the result be published here?

Thank you for your interest. We are considering using the overall findings from this study in a future scientific publication. Since the survey is anonymous, individual responses will not be published in their original form. However, we hope to share the aggregated results in the future once the full analysis has been completed.

 

[voodooless]

It means a new method to digitize information.
The most classic method is PCM. However, one can devise their own method to digitize signals via new forms of ADCs.

Those files being compared are PCM, right? So what up with that? What is that novel method? What is the advantage/disadvantage?

 

[USF-Audio-2026]

Those files being compared are PCM, right? So what up with that? What is that novel method? What is the advantage/disadvantage?

Yes — the playback files are standard PCM audio. The novelty is not in the playback format, but in the acquisition pipeline that produced them. Conventional ADCs clip when the signal exceeds range, whereas the alternative method uses a non-linear folding operation and then reconstructs the signal computationally. You can think of recovering signals purely from quantization noise or error. The potential benefit is better recovery in difficult high-dynamic-range settings for the same bit-depths.

 

[voodooless]

Yes — the playback files are standard PCM audio. The novelty is not in the playback format, but in the acquisition pipeline that produced them. Conventional ADCs clip when the signal exceeds range, whereas the alternative method uses a non-linear folding operation and then reconstructs the signal computationally. You can think of recovering signals purely from quantization noise or error. The potential benefit is better recovery in difficult high-dynamic-range settings for the same bit-depths.

Thanks! I had a quick look into this, and it’s quite interesting! I’m not sure how useful this is for audio though, especially if you have to scale back your infinite DR signal to the fixed DR of a normal PCM signal.

 

[Timcognito]

a non-linear folding operation and then reconstructs the signal computationally

Sounds similar to what they claimed with MQA. IMO the folding was the most interesting thing that they did, but no one seemed to give that part a second thought.

 

[USF-Audio-2026]

Thanks! I had a quick look into this, and it’s quite interesting! I’m not sure how useful this is for audio though, especially if you have to scale back your infinite DR signal to the fixed DR of a normal PCM signal.

You can find an overview of the method here: https://www.siam.org/publications/s...imited-sensing-when-noise-becomes-the-signal/
In the audio setting, we have developed and validated the hardware pipeline in our lab. Our results to date indicate a clear advantage in mitigating the clipping problem relative to conventional ADC-based acquisition, without sacrificing digital resolution.

 

[USF-Audio-2026]

Sounds similar to what they claimed with MQA. IMO the folding was the most interesting thing that they did, but no one seemed to give that part a second thought.

The folding of amplitudes is being considered here, as shown below.

 

[voodooless]

Sounds similar to what they claimed with MQA. IMO the folding was the most interesting thing that they did, but no one seemed to give that part a second thought.

That’s “folding” in the frequency domain. That’s a totally different thing. And really, “folding” is kind of a big word here, because what they do if remove half of the samples, and compress the removed half into the lowest LSBs of the remainder.

 

[LTig]

Made the test.

Spoiler: Results

All the samples are rather low sound quality, but one always distorts horrible. IMV it makes no sense to add such a sample, as the test actually results in an AB comparison.

 

[Victorr_]

My impressions. No sample scores more than 40 points (from 10 to 40). Those that are distorted - definitely 0 points.