Unforgettable Audio Recording Snippets

In the Unforgettable App, audio can be recorded in both raw (WAV) and obfuscated formats. By default only obfuscated audio is retained to protect both your and other people’s privacy, so you will need to go to the Settings panel in the App and enable raw audio if you wish to store it. Raw audio segments are 6 seconds in duration and are taken every 10 minutes by default. Note though that the ethics and legality of recording raw audio varies by context and jurisdiction, so be certain you have considered these issues before enabling raw audio recording.

The audio segments that are recorded by default are only 500 milliseconds in duration and are retained as Mel-Frequency Cepstral Coefficients (MFCC)i. Again they are recorded every 10 minutes. These segments are sufficient to infer some aspects of the ambient audio environment - like the presence of voice or traffic noise - but cannot be used to reconstruct conversation.

To get better understanding of how the audio is obfuscated, listen to the following obfuscated clip:

Can you understand what was said? The previous snippet was generated from this raw audio clip:

Can you tell which part?

The retention of the obfuscated audio is only useful if it allows you to infer properties of the environment. Listen to the following clip:

You should be able to tell that this clip was generated from a different noise environment than the first obfuscated clip even though you are unlikely to be able to tell what was being said. For reference, this is the raw audio clip from which it was generated:

Audio information can be informative, but it also raises ethical and legal concerns. The recording of obfuscated audio should eliminate most of these concerns, while still providing useful information. At the same time, the ability to enable raw audio gives you the power to do that when you are confident you are in an appropriate environment.

We value your feedback. Do you think we have struck the right balance?