We present the results of our audio flow computation on various phonetic transitions. In each case, the first and last envelope of a phonetic transition was extracted and submitted for audio flow computation. We present the results from the three algorithms in our paper: direct audio flow, multiresolution audio flow (with a level 6 Gaussian pyramid), and graph flow (with flow concatenation across paths extracted from a 20 sec corpus).
In all the images below, our audio flow results are presented as the red lines depicting the correspondence between the first envelope (on the bottom) and the last envelope (on the top).
Direct Audio Flow | Multiresolution Audio Flow | Graph Multiresolution Audio Flow | |
/ay/ | Direct | Multires | Graph |
/oy/ | Direct | Multires | Graph |
/ow/ | Direct | Multires | Graph |
/aw/ | Direct | Multires | Graph |
/weh/ | Direct | Multires | Graph |
/aar/ | Direct | Multires | Graph |
/ehl/ | Direct | Multires | Graph |
/mao/ | Direct | Multires | Graph |
/ahn/ | Direct | Multires | Graph |
/yuw/ | Direct | Multires | Graph |
/ey/ | Direct | Multires | Graph |