AudioGPT/NeuralSeq/usr/__pycache__/diffspeech_task.cpython-37.pyc

B
<00>Xd<><00>@s<>dd
lZdd
l
Z
ddlmZ
ddlmZ
ddlmZ
ddlm	Z	
ddl
mZmZ
ddl
mZ
dd	lmZ
d
dd<0C>i
ZGd
d<0E>de	<09>Zd
S)<0F>N)
<01>hparams<6D>
)
<01>DiffNet)
<01>GaussianDiffusion)
<01>
DiffFsTask)<02>get_vocoder_cls<6C>BaseVocoder)
<01>	denorm_f0)
<01>FastSpeechDataset<65>wavenetc

Cst|d
<00>
S)N<>audio_num_mel_bins)
r)
<01>hp<68>r<00>V/mnt/sdc/hongzhiqing/code/audio_chatgpt/text_to_sing/DiffSinger/usr/diffspeech_task.py<70><lambda>
<00>rcsHeZ
dZ<02>f
d
d<02>Zdd<04>Zdd<06>Zddd	<09>
Zd
d<0B>Zdd
d<0E>
Z<08>Z	S)<11>DiffSpeechTaskc

s$tt
|<00><02><02>
t|_tt<06>
<01>|_dS)
N)<08>superr<00>__init__r
<00>dataset_clsrr<00>vocoder)
<01>self)
<01>	__class__rrrs


zDiffSpeechTask.__init__c

Cs<>td
}
t
|j|
ttdt<00>
tdtdtdtdtdd<08>|_td	d
krjtj|jjtd	ddd
<0A>
x&|jj<07><08>D]\}}d|krxd|_	qxWdS)Nr<00>diff_decoder_type<70>	timesteps<70>K_step<65>diff_loss_type<70>spec_min<69>spec_max)<08>
phone_encoder<65>out_dims<6D>
denoise_fnrr<00>	loss_typerr<00>fs2_ckpt<70><00>modelT)
<01>strictZ	predictorF)
rrr<00>
DIFF_DECODERSr%<00>utils<6C>	load_ckpt<70>fs2<73>named_parameters<72>
requires_grad)r<00>mel_bins<6E>
k<>
vrrr<00>build_tts_models










zDiffSpeechTask.build_tts_modelcCs@tj
jtd
d<02>|
<01><04><00>tdtdtdftdd<07>|_}|S)Nc


Ss|jS)
N)
r,)
<01>
prrrr*rz0DiffSpeechTask.build_optimizer.<locals>.<lambda><3E>lrZoptimizer_adam_beta1Zoptimizer_adam_beta2<61>weight_decay)r2<00>betasr3)<07>torch<63>optim<69>AdamW<6D>filter<65>
parametersr<00>	optimizer)rr%r:rrr<00>build_optimizer(s





zDiffSpeechTask.build_optimizerFc
Cs
|d
}|d}|d}|d}|d}	|d}
tdsB|<02>
d<08>
n|<02>
d	<09>
}td
dkr<>|d}|d
}
|d}|
<01>||
||<07>|d<}|
||||||	|
|d<10>}i}d|kr<>|d|d<|j|d|||d<14>
tdr<>|<00>|||<10>
tdr<>|<00>|d|
|<10>
|<03>
s|S||fSdS)N<>
txt_tokens<EFBFBD>mels<6C>mel2ph<70>f0<66>uv<75>energy<67>
use_spk_id<EFBFBD>	spk_embed<65>spk_ids<64>
pitch_type<EFBFBD>cwt<77>cwt_spec<65>f0_mean<61>f0_std<74>f0_cwt)r>rC<00>ref_melsr?r@rA<00>infer<65>	diff_loss<73>mel<65>dur)
<01>losses<65>use_pitch_embed<65>use_energy_embed<65>energy_pred)r<00>get<65>cwt2f0_norm<72>add_dur_loss<73>add_pitch_loss<73>add_energy_loss)rr%<00>sample<6C>
return_outputrLr<<00>targetr>r?r@rArCrGrHrI<00>outputrPrrr<00>	run_model0s2



















zDiffSpeechTask.run_modelc
Cs
i}|
d
}|
d}tds&|
<01>
d<04>
n|
<01>
d<05>
}|
d}|
d}|
d}	i|d	<|j|j|
d
dd<0C>\|d	<}
t|d	<00><05><00>
|d
<|
d|d<t<06>|<03>
}|tdk<00>
r|j|||||	|dd
d<10>}
t|
d|
dt<00>}|j	||
d|
dd
||
<EFBFBD>
d<13>
d<14>
|<00>
||
d|
d<00>
|S)Nr<rArBrCrDr>r?r@rPTF)rZrL<00>
total_loss<EFBFBD>nsamples<65>num_valid_plots)rCr>r?r@rArKrLr=<00>mel_out<75>	f0_denorm)<03>is_mel<65>gt_f0r?)rrTr]r%<00>sum<75>valuesr(<00>tensors_to_scalarsr	<00>plot_wav<61>plot_mel)rrY<00>	batch_idx<64>outputsr<rArCr>r?r@<00>	model_outrdrrr<00>validation_stepPs&












$

zDiffSpeechTask.validation_stepNcCs<>|d
<00><00><00>
<01>}|d
<00><00><00>
<01>}|d
<00><00><00>
<01>}|d
<00><00><00>
<01>}|rd|jj||d<02>}|jj||d<02>}|jjjd|
<01><00>|td|jd<05>
|jjjd|
<01><00>|td|jd<05>
dS)Nr
)
r?<00>gt_<74>audio_sample_rate)<02>sample_rate<74>global_stepZwav_)	<09>cpu<70>numpyr<00>spec2wav<61>loggerZ
experimentZ	add_audiorrq)rrjZgt_wav<61>wav_outrcrdr?<00>namerrrrhps







"
zDiffSpeechTask.plot_wav)FF)FNNN)
<EFBFBD>__name__<5F>
__module__<EFBFBD>__qualname__rr0r;r]rmrh<00>
__classcell__rr)
rrrs

  r)r5r(<00>
utils.hparamsr<00>diff.netr<00>diff.shallow_diffusion_ttsr<00>taskr<00>vocoders.base_vocoderrr<00>utils.pitch_utilsr	<00>tasks.tts.fs2_utilsr
r'rrrrr<00><module>
s