| 摘要 |
A technique for extracting the onsets and offsets from each channel of bandpassed sound and for temporally and tonotopically clustering these onsets and offsets using a neural network. The onset and offset maps so produced are used to segment the sound. The segments so produced may be analysed one by one for separate interpretation. Each segment can then be identified as belonging to one source or not. The arrangement shows parallel processing of the positive and negative going parts of the onset/offset filtered signal. The neural networks 5a, 5b sharpen the onsets, offsets and cluster them across the frequency bands.
|
