AFP (Audio FingerPrinting) is an effective paradigm for music retrieval, which is also knowns as "query by example". Based on a given noisy recording of 8 seconds, our AFP system can identify the intended piece of music is less than a second, with an retrieval accuracy of 95%. AFP is also very power in content recognition for any media that contains audio, including video. Common application scenarios of AFP includes:
- Detection of music/audio/video copy right violation, for media upload in social network, etc.
- Automatic royalty computation for music broadcast/streaming service, etc.
- Second-screen applications which detect video segments for product placement or embedded advertising.
- TV or commercial ratings for product promotion activities.
The deployment of a AFP system involves the following steps:
- Collect a set of music/audio/video the database
- Extract features of these audio recordings.
- Put the features into a compact hash table for easy comparison.
We are flexible in promoting our AFP technology. In particular, we can deploy an AFP system based on a common PC system of 128G RAM and a GPU card of 1080Ti, to host a database of 100M pieces of music for recognition.