Minimum duration for successful identification (CID)

[mmuser]

I wonder what the minimum duration of a music sample needed for Content ID to make a successful identification is?

The reason I ask is, that besides the Main Tracks, I also uploaded some alternates (like Bed Acoustic) to Content ID. To my concern, I noticed that in a few of these cases (fortunately, only a few), for a few seconds (usually up to 3-4 seconds), there might be very few VST instruments remaining.

In two instances, there was just one instrument (an acoustic guitar playing an automatic strum). Additionally, sometimes 3-4 VST instruments (including 1-2 Arpeg) might play for longer periods.

Or in another instance, there was a vst brass playing full chords by itself for 8-9 seconds (where you play one note and it performs a chord).

How likely is it for a false claim to occur because of this?

Generally, I don’t compose with ARPs and automated phrases, but some older tracks concern me.

[Art Munson]

YouTube ContentID can recognize short clips. But, I would imagine if it’s too simplistic, or similar, you could get false positives or negatives.

[mmuser]

Thanks, Art! What I don’t know is how “sensitive” their system is. For example, in the most concerning scenario:
of a VST guitar playing a chord with automated strumming on its own for 3-4 seconds. For a false claim to be an issue, I would imagine another composer would have to use the exact same chord, at the same tempo, and possibly with the same settings. Unless the system also detects similar audio. That is, if two songs have the same vst guitar sound and the same chord, I wonder if the system could ignore a slightly different tempo and consider them the same?

In any case, I shouldn’t have uploaded those alternate versions where the orchestrations were “bare” in some spots (fortunately, there were only a few).

By the way, here is a general response I received from GPT on this topic (What reassures me a bit is the last sentence.):

For most modern Content ID systems:

Typical minimum duration: 3-5 seconds
Many systems can identify a song with just a few seconds of audio.

Optimal duration: 10-15 seconds
This length usually provides a good balance between quick identification and accuracy.

Variable factors:
Audio quality
Distinctiveness of the music segment
Background noise
System’s algorithm efficiency

Platform specifics:
YouTube’s Content ID system, for example, can often identify music in as little as 5 seconds, but may use longer segments for increased accuracy.
False positives:
Very short clips (1-2 seconds) might lead to more false positives, which is why most systems use slightly longer samples.

It’s worth noting that while a system might be able to identify a song in just a few seconds, many platforms choose to analyze longer segments to ensure accuracy, especially when making copyright-related decisions.

[Author]

Posts