Energy-based audio pattern recognition with weighting of energy matches

摘要

An energy-based pattern recognition algorithm receives the input frames of an audio signal and a test frame sequence and returns a best match in the audio signal to the given test frame sequence. The energy of each input frame is computed and input frames for which the energy is both within a predetermined degree of closeness to the local maximum energy within the test frame sequence and a local maximum within a respective neighborhood of adjacent frames are identified as probable matches. The difference between overall energy for frames neighboring the remaining probable matches and the test frame sequence is computed as a percentage. The best match is selected based on a weighted combination of difference between local maximum energies and minimum percent deviation in overall energy. Local signal characteristic matching may be employed, with weighting, to refine matching.