Identification/Matching
How does the identification process work?
A Pex provided Software Development Kit (SDK) enables clients to locally fingerprint content and return the fingerprints for identification. Local fingerprinting is both faster and safer for clients, as original content does not need to be shared with Pex, and Pex cannot restore original content from fingerprints. Fingerprints are compact representations of content that process quickly, which reduces costs and enables real-time results.
Pex returns match results to clients via the SDK. The results detail whether or not there was a match found, as well as the timecodes of the identified assets. If searching the Pex Registry, additional asset metadata such as song title, artist and rightsholder is returned.
What content can be submitted for fingerprinting?
Clients can create an audio or melody/phonetic fingerprint from any standard media file. For video files, clients will not need to extract audio from video before creating a fingerprint.
How quickly can content be fingerprinted and identified?
Pex technology can create fingerprints, identify matches, and return results in real time (five seconds or less).
What is the difference between an audio search and a melody/phonetic search?
An audio search will return any matches we identify to the sound recording itself. This includes matches where the sound recording has been modified either by pitch and/or speed.
A melody/phonetic search will return any matches to the same underlying composition or lyrics, such as cover songs or live versions of a song. This includes matches where the melody has been transposed.
Can Pex still identify matches if the content has been modified?
Pex’s technology is finely tuned to make identifications even if the input file has been modified in a variety of manners. Below are some of the modifications we are able to handle:
Pitch: Robustly match audio content with pitch between -900 and +1200 cents (between -9 and +12 semitones). In many cases, Pex can match plus or minus 1250 cents (12.5 semitones).
Speed: Robustly match audio content with 50-150% of the original speed. In many cases, Pex can match 50-200%.
Pitch + speed: Robustly match pitch + speed together at 66-150% of the original. In many cases, Pex can match 50%-200% of the original.
Pex can also robustly match
Very short samples (1s)
Mashups
Multi-track mixes
Remixes
Can content be fingerprinted and identified during a live stream?
Yes. For live streams, content can be sent to Pex in short segments, which are matched in real time with results returned for every segment.
For example, a 30 minute live stream can be fingerprinted and queried in 60 second intervals with results returned in 5 seconds ot less.
What is included in the match results when searching against the Pex Registry?
For identified matches, Pex returns:
Asset Metadata, including song title, artist, and any available ownership information such as record label
Match Details, including matched segments (e.g. 0:00-0:15 of Song X was identified in 1:25-1:40 of Audio X) and any modification info (e.g. Song X
How can I use the confidence score that is returned in a search response?
The confidence score reflects the system's certainty about each segment decision, providing a relative measure of certainty for each match, ranging from 85 to 100. While the score indicates how closely a segment matches, it’s intended to aid in QA prioritization rather than serve as a strict measure of match accuracy or consistency.
Using the Confidence Score The main purpose of the confidence score is to rank and sort results, helping prioritize QA efforts. Lower scores may suggest a slightly higher chance of false positives, making them useful for guiding manual review.
Limitations
Not an accuracy measure: This score should not be used for automated filtering or as an absolute accuracy indicator.
Subject to change: Confidence values may vary with each SDK release due to algorithm updates, so they don’t provide a stable filtering threshold.
Last updated