Query-based Audio Matching

With query-based matching Fraunhofer IDMT offers a robust solution to detect a given audio query within a stream or file – even under noisy conditions or with a very short query. The software extracts acoustic fingerprints from query and reference files or streams, matches them against each other and shows the exact position of the overlapping content. The tool can be used to optimize broadcast and ad-monitoring or duplicate detection in digital archives.

Robustness to noise and interfering signals

Query-based matching detects whether a given query signal like an ad or a song segment is used in a TV and radio broadcast – even if the sample file is noisy due to mobile capturing and air transmission or if the audio material is used in a mix and other signals are present within the stream. No appearance is going to be missed even under difficult conditions, so tedious manual content tracking is no longer necessary.

Automatic detection of any audio content

Fraunhofer IDMT’s query-based matching is not limited to music. It works with audio signals of all types like speech, music, environmental sounds and noises, signals with mixed content, and noise-like signals. The software can therefore be used for a wide range of applications and tasks.

High accuracy

The software recognizes even very short queries of 3 to 4 seconds with a high accuracy. For this reason, automatic detection can also be used reliably to detect very short audio queries like station IDs or audio brands.

Application Scenarios

  • Ad monitoring: Automatic identification and tracking of ads in radio or TV programs
  • Content tracking: Metadata propagation and rights clearance
  • Archive Cleanup: Detection of duplicates and content reuse in digital archives
  • Broadcast monitoring and radio or TV stream analysis

Features

  • Fingerprinting-based technology
  • Robustness to noise: detecting matches with signal-to-noise-ratio (SNR) up to -15 dB
  • Robustness to signal distortions like echoes, reflections, equalization
  • Robustness to low bitrate audio coding:  48 kbps, 80 kbps and 128 kbps
  • Detection of speech, music, environmental sounds and noises, signals with mixed content, and noise-like signals
  • Working with very short queries of 3 to 4 seconds with up to 99%  precision and recall
  • Integration in existing systems and customization on request
  • Free test version available on request

This might also be of interest

Research topic

Audio and Visual Content Analysis

Extracting meaningful data from audiovisual content

 

Research topic

Automatic Music Analysis

 

Reference project

SWR Media Services

Audio Matching Software für automatisches Werbemonitoring von SWR Radioprogrammen