Skip to content

zhu-yifang/Yifang-thesis

Repository files navigation

Yifang-thesis

  • Read all the files
  • Get all the phones
  • Get the MFCC sequences of all the phones
  • Train
  • Test
  • Get the number of each phone
  • Remove the silence and pause and do the folding
  • Build a Confusion matrix, get 50 samples for each phone
  • Export to Excel automatically
  • Try clustering or normalizing
  • Try reduced training sets
  • Investigate normalizing instances
  • Power spectra, diff MFCC features
  • Clustering training instances
  • Write up what I have

The stats of the training set: Counter({'h#': 9240, 'ix': 8642, 's': 7475, 'n': 7068, 'iy': 6953, 'tcl': 6644, 'r': 6539, 'kcl': 5859, 'l': 5801, 'ih': 5051, 'dcl': 4942, 'k': 4874, 't': 4364, 'ae': 3997, 'm': 3903, 'eh': 3853, 'z': 3773, 'ax': 3610, 'q': 3590, 'd': 3548, 'axr': 3407, 'w': 3140, 'aa': 3064, 'ao': 2940, 'dh': 2826, 'dx': 2709, 'pcl': 2644, 'p': 2588, 'ay': 2390, 'ah': 2306, 'ey': 2282, 'sh': 2238, 'gcl': 2223, 'f': 2216, 'b': 2181, 'ow': 2136, 'er': 2046, 'g': 2017, 'v': 1994, 'bcl': 1909, 'ux': 1908, 'y': 1715, 'epi': 1464, 'ng': 1330, 'jh': 1209, 'hv': 1154, 'nx': 971, 'hh': 957, 'pau': 952, 'el': 951, 'ch': 822, 'th': 751, 'aw': 729, 'en': 723, 'oy': 684, 'uw': 555, 'uh': 535, 'ax-h': 375, 'zh': 151, 'em': 124, 'eng': 38})

The stats of the testing set: Counter({'h#': 3360, 'ix': 2945, 'iy': 2710, 's': 2639, 'r': 2525, 'n': 2501, 'l': 2356, 'tcl': 2334, 'kcl': 1964, 'ih': 1709, 'dcl': 1643, 'k': 1614, 't': 1535, 'm': 1526, 'eh': 1440, 'ae': 1407, 'axr': 1383, 'ax': 1346, 'z': 1273, 'd': 1245, 'q': 1244, 'w': 1239, 'ao': 1156, 'aa': 1133, 'dh': 1053, 'pcl': 965, 'p': 957, 'dx': 940, 'f': 912, 'b': 886, 'ah': 879, 'ay': 852, 'gcl': 808, 'ey': 806, 'er': 800, 'sh': 796, 'ow': 777, 'bcl': 776, 'g': 755, 'v': 710, 'y': 634, 'ux': 580, 'epi': 536, 'ng': 414, 'pau': 391, 'jh': 372, 'hv': 369, 'nx': 360, 'hh': 356, 'el': 343, 'th': 267, 'oy': 263, 'ch': 259, 'en': 251, 'uh': 221, 'aw': 216, 'uw': 170, 'ax-h': 118, 'zh': 74, 'em': 47, 'eng': 5})

Remove h#, #h, sil, pau, epi

merge/fold (axr, er), (m, em), (n, en, nx), (ng, eng), (pcl, p), (tcl, t), (kcl, k), (bcl, b), (dcl, d), (gcl, g)

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published