Determining audio events in longer audio files

[erikparr]

Very excited to find out about the freesound-datasets project. One idea for future implementations: I think it would be useful to create a way of determining audio events within longer audio files. For example, if an audio file consists of a field-recording that contains, say, church bells from 1:15-1:25. Such audio events that stand out from the rest of the recording could be determined via amplitude spikes, or probably more sophisticated ML techniques. I would be interested in knowing if algorithms exist to achieve something like this.

If such a feature were implemented it would significantly expand the material available for the dataset.

[xavierfav]

It would something very useful indeed.
In #123 we propose to create a task for enabling the manual definition of the time regions where a specific event occurs.

It would be interesting, as you point, to have an automatic method that can already propose some regions that could be then refined by people.