def _get_inputs(self):
n_timesteps = (hp.default.duration * hp.default.sr) // hp.default.hop_length + 1
return [InputDesc(tf.float32, (None, n_timesteps, 9999), 'x_ppgs'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_mfcc), 'x_mfccs'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_fft // 2 + 1), 'y_spec'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_mels), 'y_mel'), ]
def _get_inputs(self):
n_timesteps = (hp.default.duration * hp.default.sr) // hp.default.hop_length + 1 #401
return [InputDesc(tf.float32, (None, n_timesteps, hp.default.n_mels), 'r_mel'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_fft // 2 + 1), 't_spec'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_mels), 't_mel'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_fft // 2 + 1), 'r_spec')]
def _get_inputs(self):
length = hp.signal.duration * hp.signal.sr
length_spec = length // hp.signal.hop_length + 1
return [
InputDesc(tf.float32, (None, length), 'wav'),
InputDesc(tf.float32, (None, length_spec, hp.signal.n_mels), 'x'),
InputDesc(tf.int32, (None, ), 'speaker_id')
]
def _get_inputs(self):
return [
InputDesc(tf.float32, (None, None, hp.default.n_mfcc), 'x_mfccs'),
InputDesc(tf.int32, (
None,
None,
), 'y_ppgs')
]
def _get_inputs(self):
length_melspec = hp.signal.length // hp.signal.hop_length + 1
return [
InputDesc(tf.float32, (None, hp.signal.length), 'wav'),
InputDesc(tf.float32, (None, length_melspec, hp.signal.n_mels),
'melspec'),
InputDesc(tf.float32, (None, 2), 'labels')
]
def _get_inputs(self):
return [
InputDesc(tf.float32, (None, 6), 'state'),
InputDesc(tf.float32, (None, 2), 'action'),
InputDesc(tf.float32, (None, ), 'futurereward'),
InputDesc(tf.float32, (None, ), 'advantage'),
# InputDesc(tf.float32, (None,), 'action_prob'),
]
def _get_inputs(self):
return [
InputDesc(tf.float32, (None, None, hp.default.n_mfcc), 'x_mfccs'),
InputDesc(tf.float32, (None, None, hp.default.n_fft // 2 + 1),
'y_spec'),
InputDesc(tf.float32, (None, None, hp.default.n_mels), 'y_mel'),
]
def _get_inputs(self):
n_timesteps = (hp.default.duration *
hp.default.sr) // hp.default.hop_length + 1
# timestep의 갯수, 전체 길이를 hop length 만큼 나눈 것.
# STFT 변환 시, 전체 오디오에서 생성되는 총 window 개수와 동일
# 그래프에 입력하는 entry point에 관한 메타데이터를 생성한다.
# 이러한 메타데이터는 나중에 placeholder을 만들거나
# 다른 타입의 입력값을 만드는데 쓰일 수 있다.
# placeholder : 데이터의 형태만 지정한 뒤, 실제 입력은 실행 단계에서 받을 수 있는 텐서를 뜻한다.
return [
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_mfcc),
'x_mfccs'),
InputDesc(tf.float32,
(None, n_timesteps, hp.default.n_fft // 2 + 1),
'y_spec'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_mels),
'y_mel'),
]
def _get_inputs(self):
#n_timesteps = (hp.default.duration * hp.default.sr) // hp.default.hop_length + 1
n_timesteps = (hp.default.duration * hp.default.sr) // hp.default.hop_length
return [InputDesc(tf.float32, (None, n_timesteps, hp.default.n_mels), 'y_mel'),
InputDesc(tf.float32, (None, n_timesteps, hp.default.n_ppgs), 'ppgs'),]
def _get_inputs(self):
return [
InputDesc(tf.float32, (None, self.length, 1), 'wav'), # (n, t)
InputDesc(tf.float32, (None, self.t_mel, hp.signal.n_mels),
'melspec')
] # (n, t_mel, n_mel)
