nnmnkwii.metrics.lf0_mean_squared_error¶
-
nnmnkwii.metrics.
lf0_mean_squared_error
(src_f0, src_vuv, tgt_f0, tgt_vuv, lengths=None, linear_domain=False)[source]¶ Mean squared error (MSE) for log-F0 sequences.
MSE is computed for voiced segments.
Parameters: - src_f0 (ndarray) – Input log-F0 sequences, shape can be either of
(
T
,), (B x T
) or (B x T x 1
). Both Numpy and torch arrays are supported. - src_vuv (ndarray) – Input voiced/unvoiced flag array, shape can be either
of (
T
, ), (B x T
) or (B x T x 1
). - tgt_f0 (ndarray) – Target log-F0 sequences, shape can be either of
(
T
,), (B x T
) or (B x T x 1
). Both Numpy and torch arrays are supported. - tgt_vuv (ndarray) – Target voiced/unvoiced flag array, shape can be either
of (
T
, ), (B x T
) or (B x T x 1
). - lengths (list) – Lengths of padded inputs. This should only be specified if you give mini-batch inputs.
- linear_domain (bool) – Whether computes MSE on linear frequecy domain or log-frequency domain.
Returns: mean squared error.
Return type: - src_f0 (ndarray) – Input log-F0 sequences, shape can be either of
(