f0eval

class diffsptk.F0Evaluation(reduction: str = 'mean', out_format: str = 'f0-rmse-cent')

See this page for details. Note that the gradients cannot be calculated if the output format is related to voiced/unvoiced decision.

Parameters:

reduction[‘none’, ‘mean’, ‘sum’]

The reduction type.

out_format[‘f0-rmse-hz’, ‘f0-rmse-cent’, ‘f0-rmse-semitone’, ‘vuv-error-rate’, ‘vuv-error-percent’, ‘vuv-macro-f1-score’]

The output format.

forward(x: Tensor, y: Tensor) → Tensor

Calculate F0 metric.

Parameters:

xTensor [shape=(…, N)]

The input F0 in Hz.

yTensor [shape=(…, N)]

The target F0 in Hz.

Returns:

outTensor [shape=(…,) or scalar]

The F0 metric.

diffsptk.functional.f0eval(x: Tensor, y: Tensor, reduction: str = 'mean', out_format: str = 'f0-rmse-cent') → Tensor

Calculate F0 metric.

Parameters:

xTensor [shape=(…, N)]

The input F0 in Hz.

yTensor [shape=(…, N)]

The target F0 in Hz.

reduction[‘none’, ‘mean’, ‘sum’]

The reduction type.

out_format[‘f0-rmse-hz’, ‘f0-rmse-cent’, ‘f0-rmse-semitone’, ‘vuv-error-rate’, ‘vuv-error-percent’, ‘vuv-macro-f1-score’]

The output format.

Returns:

outTensor [shape=(…,) or scalar]

The F0 metric.

See also

rmse