The extraction of information from audio files has applications in the detection and classification of musical segments, timbres and biometrics, and the appearance of anomalies in an audio track.
Analysis of spectrograms with neural networks transcends traditional tape analysis based on transcription to text, enabling nuances to be extracted which may help detect important information in the audio file.
A leading company in the analysis of advertising information in online and off-line media has an extensive team of people listening to radio recordings to detect advertising mentions.
A semi-automatic analysis tool assists the process with an accuracy of close to 80%, insufficient to free up human verification and extraction work.
Foqum’s spectrogram analysis tool achieved an accuracy over 87% in the detection of important segments with a training set below 5% of the company's entire database, including advertising mentions in different languages.
The scalability of the tool to the use of all the channels and recording times has an accuracy over 90% and boosts the efficiency of the human classification and extraction teams.