Hacker News Reader

Comment by Sai_ | Hacker News Reader

Interesting.

Mind if I ask how a sound recorded by two different sensors can be identified as the same original sound if the loudness is different? Unless the frequency of loud sounds is less than the time taken by the sounds to reach the sensors, wouldn’t it be hard to pick up two sounds spread apart by say a second but know that they are the same sound?

For example, if I’m on the phone with a friend during a thunderstorm and we are on opposite sides of the city, will we hear and be able to identify the same thunderclap?

Does sound have usable distinguishing features other than just loudness?

View on HN

Sounds have pretty characteristic shapes, in both waveform (amplitude/time) and frequency spectra/time. So much so that when I'm doing an editing job and I know the source material well, I can often do tidy-up edits for 30 seconds' worth of audio by eye and be confident it will play back correctly. So you can do a lot of preliminary filtering by looking out for the distinctive shapes of percussive noises, as opposed to other loud things like vehicle horns, motors, or music.

Back in the shotspotter example, your microphones are in fixed positions, so you can figure out in advance what the echoes are going to be for any given originating sound, and do a reasonable job of triangulation with 3 sensors. You also have some other givens, eg it's reasonable to assume that the location of any actual gunshot is 1-2 meters above ground level; this means you'll be slightly wrong about rooftop snipers/shootouts but those are vanishingly rare compared to street shootings. I don't think anyone claims the product to be reliable for detecting shootings in building interiors, where you'd be competing with televisions and other false positives.

As I mentioned earlier, a product like this can get away with a relatively low success rate, as long as it's slightly better than the information that comes in from 911 calls and the rate of false positives if either tolerably low or negligently unaccounted for.

> Does sound have usable distinguishing features other than just loudness?

Than raw, peak amplitude? yes.

Frequency and amplitude variation over time will be roughly consistent in a linear system (i.e. the sound will get louder and quieter over time). If you have a stable, well-qualified listening system (such as, for example, a set of fixed recording locations in an environment that is acoustically static), you can build a map of how sound sources map to each recorder, and use principles of directed arrays to get very accurate location projections.

To use your example, yes it would be difficult for you and a single friend to confirm that you heard the same sound, but you could come up with a probability that if you heard a sound with the same characteristics with a certain time difference and loudness difference, it is likely to have been from a certain location. If, however, you have 4 friends, it gets much less likely that all 4 heard a different sounds calibrated exactly to seem like they came from a specific source when the data is combined.

Reply on news.ycombinator.com