AudioGPT/NeuralSeq/modules/diff/__pycache__/candidate_decoder.cpython-37.pyc

B

9d<><00>@s<>dd
lm
Z

ddlZddlmZ
ddlmZ
ddlZddlm	Z	
ddl
mZ
ejZGdd<08>dej
<0A>Zd	d
<EFBFBD>ZGdd<0C>de
<01>ZdS)
<0A>)
<01>FastspeechDecoderN)
<01>
functional)
<01>hparams<6D>
)
<01>Mishcs$eZ
dZ<02>f
d
d<02>Zdd<04>Z<04>ZS)<05>SinusoidalPosEmbcst<00><00>
<01>
|
|_dS)
N)<03>super<65>__init__<5F>dim)<02>selfr
)
<01>	__class__<5F><00>./usr/diff/candidate_decoder.pyr	s


zSinusoidalPosEmb.__init__cCsz|
j}|j
d
}t<02>d<02>
|d}t<04>tj||d<04>|<00>
}|
dd<00>df|ddd<00>f}tj|<04><08>|<04>	<09>fdd<06>}|S)N<>i'r)
<01>device<63><65><EFBFBD><EFBFBD><EFBFBD>)
r
)
rr
<00>math<74>log<6F>torch<63>exp<78>arange<67>cat<61>sin<69>cos)r<00>
xr<00>half_dim<69>embr
r
r<00>forwards





 

zSinusoidalPosEmb.forward)<06>__name__<5F>
__module__<EFBFBD>__qualname__r	r<00>
__classcell__r
r
)
rrrs
rcOstj
||
<01>
}tj<02>|j<04>

|S)
N)<05>nn<6E>Conv1d<31>init<69>kaiming_normal_<6C>weight)<03>args<67>kwargs<67>layerr
r
rr#s


r#cs(eZ
dZd<07>f
dd<03>	Zddd<06>
Z<04>ZS)	<09>FFTNcs<>t<00>j
|
|||d
<01>
td}ttd|d<04>|_t|<05>
|_t<07>t<07>	||d<00>t
<EFBFBD>t<07>	|d|<05><02>|_t	tdddd	<09>|_t	td||td<00>|_
dS)
N)
<01>	num_heads<64>residual_channels<6C>audio_num_mel_binsr<00><00>hidden_size<7A>PT)
<01>bias)rr	rr#<00>input_projectionr<00>diffusion_embeddingr"<00>
Sequential<EFBFBD>Linearr<00>mlp<6C>get_mel_out<75>get_decode_inp)rr/<00>
num_layers<EFBFBD>kernel_sizer+r
)
rr
rr	$s










zFFT.__init__FcCs<>
|
d
d
<01>df}|<00>|<07>
<01>
dddg<03>
}|<00>|<02>
}|<00>|<02>
}|<03>
dddg<03>
}|jd}|d
d
<01>d
d
d
<01>f}	|	<09>d|dg<03>
}	tj|||	gdd<06>}
|<00>|
<EFBFBD>
}
|
}|d
kr<>|<07>	<09><00>
d<05>
<01>d<02>
jn|}d|<04>
dd<04><02><0E>d
d
<01>d
d
<01>d
f}|j<0F>
r|j|<00>|d<00>
}||}tj||j|jd<08>}|<07>
dd<04>|}g}
x,|jD]"}||||d	<09>|}|
<0A>|<07>

<00>
q8W|j<17>
rt|<00>|<07>
|}|<06>
r<>t<06>|
d<02>}|<07>
dd<03>}n|<07>
dd<04>}|<00>|<07>
<01>
dddg<03>
}|d
d
<01>d
d
d
<01>d
d
<01>fS)
z<EFBFBD>
        :param spec: [B, 1, 80, T]
        :param diffusion_step: [B, 1]
        :param cond: [B, M, T]
        :return:
        Nr
rrr)
r
).r
)<02>
p<>training)<02>encoder_padding_mask<73>	attn_mask)r2<00>permuter3r6<00>shape<70>repeatrrr8<00>abs<62>sum<75>eq<65>data<74>	transpose<73>float<61>
use_pos_embed<65>pos_embed_alpha<68>embed_positions<6E>
F<>dropoutr<<00>layers<72>append<6E>
use_last_norm<72>
layer_norm<EFBFBD>stackr7)r<00>spec<65>diffusion_step<65>cond<6E>padding_maskr><00>return_hiddensr<00>seq_lenZ
time_embed<EFBFBD>decoder_inp<6E>nonpadding_mask_TB<54>	positions<6E>hiddensr)r
r
rr2s<











"
&













zFFT.forward)NNNN)NNF)rrr r	rr!r
r
)
rrr*#s
r*)<11>modules.fastspeech.tts_modulesrr<00>torch.nnrrKr"r<00>
utils.hparamsr<00>	diffusionrr5<00>Modulerr#r*r
r
r
r<00><module>
s