About

The ATC Labs music matching API is designed for identifying the similarity or match between original and somewhat modified/processed/dilated/truncated music currently found in broadcast database (henceforth “Processed audio”). This technology is beneficial in audio database management for machine learning and audio AI processing.

In Audio broadcasting industry huge database management is required. One of the issues faced is similar type of audio present in the database. Hence, sometimes it is desirable to search for a music signal from a large database which includes similar music but with the additional constraints such as

• Samples could be truncated, or silence is added in beginning and end.
• Time scale could be different (time dilation). Even time scale difference is variable with time.
• Different digital audio parameters such as sample rate, bit-depth and channels.
• Processed for audio enhancement, bass enhancement, equalizer applied, or loudness normalized.
• Gain or attenuation applied.
• Low pass filtered (e.g. 8Khz, 4Khz).
• Samples could be compressed at low bitrates using audio codecs such as HE-AAC, MP3/MP2 or some other techniques.
• It could be a different performance e.g., live performance on stage or recording at different studio.

Music matching technique is based on our pending patent Identifying Similarity between two tracks algorithm which utilizes core DSP and pattern recognition algorithms. Techniques included efficient time dilation estimation between two tracks, SAD and frequency coherence-based searching and matching using audio features.

API provides the following matching measurements between two audio tracks

• Degree of similarity measure between original music and processed music. It is based on Coherence computation. The scale ranges from 0 to 100, where 0 indicates no similarity and 100 indicates the audio files are identical.
• Same Signature Score. The matching score between the original and processed audio is based on perception. It provides favourable scores even if the processed audio undergoes significant processing, compression, has jitter due to dilation factors, or minor edits in the middle. Degree of similarity could be low in case of heavy processed sample compared to original, but signature score comes high. The scale ranges from 0 to 100, where 0 indicates no similarity and 100 indicates the audio files are identical.