synthesis

MLPG

morgana.viz.synthesis.MLPG(means, variances, windows=None, padding_size=0, seq_len=None)

Performs maximum-likelihood parameter generation.

Parameters

• means (np.ndarray, shape (batch_size, seq_len, feat_dim) or (seq_len, feat_dim)) – Array of means for a single, or a batch of sequences.

• variances (np.ndarray, shape (batch_size, seq_len, feat_dim) or (seq_len, feat_dim) or (feat_dim)) – Array of variances for a single, or a batch of sequences.

• windows (list[tuple[int, int, np.ndarray]]) – Windows describing the static/delta features included in the feature dimension of means and variances.

• padding_size (int) – Padding on either side of signal, used to handle the boundaries.

• seq_len (array_like, shape (batch_size)) – Sequence lengths, necessary when a batch of sequences is given, as out-of-sequence frames will be all zeros.

Returns

most_probable_trajectory – The most probable trajectory, calculated by maximum-likelihood parameter generation.

Return type

np.ndarray, shape (batch_size, seq_len, feat_dim) or (seq_len, feat_dim)