e_control not used during synthesis
lordzuko opened this issue · 0 comments
lordzuko commented
The e_control
parameter fixed in the , is not correctly used in work/FastSpeech2/model/modules.py
-> VarianceAdaptor.forward
function.
Currently:
if self.energy_feature_level == "phoneme_level":
energy_prediction, energy_embedding = self.get_energy_embedding(
x, energy_target, src_mask, p_control
)
x = x + energy_embedding
and
if self.energy_feature_level == "frame_level":
energy_prediction, energy_embedding = self.get_energy_embedding(
x, energy_target, mel_mask, p_control
)
x = x + energy_embedding
Fix:
if self.energy_feature_level == "phoneme_level":
energy_prediction, energy_embedding = self.get_energy_embedding(
x, energy_target, src_mask, e_control
)
x = x + energy_embedding
and
if self.energy_feature_level == "frame_level":
energy_prediction, energy_embedding = self.get_energy_embedding(
x, energy_target, mel_mask, e_control
)
x = x + energy_embedding