AUDIO FINGERPRINT EXTRACTION AND AUDIO RECOGNITION USING SAID FINGERPRINTS

摘要

Methods and a computer-readable storage device for generating fingerprints from an audio file, and for identifying whether a query audio signal and a reference audio signal share similar audio content information, even when the query audio signal or the reference audio signal (or both) have undergone obfuscations, such as changes in pitch, changes in tempo, filtering, or the addition of background noise. A frequency representation of the query audio file is generated and a fingerprint is generated for one or more areas of data points by applying a trained neural network onto each said selected area of data points, thereby converting the area of data points into a vector in a metric space. At least one of said generated query fingerprints is compared to at least one reference fingerprint accessible by a database and generated from a set of reference audio files, by calculating the distance between said at least one generated query fingerprint and at least one reference fingerprint, using a specified distance metric. A reference audio file having associated reference fingerprints which upon said measurements of distance have produced at least one associated distance satisfying a predetermined threshold is identified.