Usage: compute-mfcc-feats [options...] <wav-rspecifier> <feats-wspecifier>
"""
po = ParseOptions(usage)
mfcc_opts = MfccOptions()
mfcc_opts.register(po)
po.register_bool(
"subtract-mean", False, "Subtract mean of each feature"
"file [CMS]; not recommended to do it this way.")
po.register_float(
"vtln-warp", 1.0, "Vtln warp factor (only applicable "
"if vtln-map not specified)")
po.register_str(
"vtln-map", "", "Map from utterance or speaker-id to "
"vtln warp factor (rspecifier)")
po.register_str(
"utt2spk", "", "Utterance to speaker-id map rspecifier"
"(if doing VTLN and you have warps per speaker)")
po.register_int(
"channel", -1, "Channel to extract (-1 -> expect mono, "
"0 -> left, 1 -> right)")
po.register_float(
"min-duration", 0.0, "Minimum duration of segments "
"to process (in seconds).")
opts = po.parse_args()
if (po.num_args() != 2):
po.print_usage()
usage = """Decode features using GMM-based model.
Usage: gmm-decode-faster.py [options] model-in fst-in features-rspecifier
words-wspecifier [alignments-wspecifier [lattice-wspecifier]]
Note: lattices, if output, will just be linear sequences;
use gmm-latgen-faster if you want "real" lattices.
"""
po = ParseOptions(usage)
decoder_opts = FasterDecoderOptions()
decoder_opts.register(po, True)
po.register_float("acoustic-scale", 0.1,
"Scaling factor for acoustic likelihoods")
po.register_bool("allow-partial", True,
"Produce output even when final state was not reached")
po.register_str("word-symbol-table", "",
"Symbol table for words [for debug output]");
opts = po.parse_args()
if po.num_args() < 4 or po.num_args() > 6:
po.print_usage()
sys.exit()
model_rxfilename = po.get_arg(1)
fst_rxfilename = po.get_arg(2)
feature_rspecifier = po.get_arg(3)
words_wspecifier = po.get_arg(4)
alignment_wspecifier = po.get_opt_arg(5)
lattice_wspecifier = po.get_opt_arg(6)
gmm_decode_faster(model_rxfilename, fst_rxfilename,
feature_rspecifier, words_wspecifier,
)
po.register_double(
"learning-rate", 200.0,
"The learning rate for t-sne is usually in the range [10.0, 1000.0]. If the"
" learning rate is too high, the data may look like a \'ball\' with any point"
" approximately equidistant from its nearest neighbors. If the learning rate"
" is too low, most points may look compressed in a dense cloud with few outliers."
" If the cost function gets stuck in a bad local minimum increasing the learning"
" rate may help. (200.0 by default)")
po.register_int(
"n-iter", 1000,
"Maximum number of iterations for the optimization. Should be at least 250. (1000 by default)"
)
po.register_str(
"distance", "euclidean", "The distance measurement between vectors. \n"
"It could be [euclidean|cityblock|mahalanobis|cosine|...]. For more options, please refer to"
" https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.pdist.html"
" (euclidean by default)")
opts = po.parse_args()
if (po.num_args() != 2):
po.print_usage()
sys.exit()
vector_rspecifier = po.get_arg(1)
vector_wspecifier = po.get_arg(2)
pdb.set_trace()
isSuccess = tsne_vector(vector_rspecifier,
vector_wspecifier,
output_dim=opts.output_dim,
perplexity=opts.perplexity,
