Method build_tts_model

usr/diffsinger_task.py:109–121

(self)

        self.dataset_cls = ShallowDiffusionOfflineDataset

    def build_tts_model(self):
        mel_bins = hparams['audio_num_mel_bins']
        self.model = OfflineGaussianDiffusion(
            phone_encoder=self.phone_encoder,
            out_dims=mel_bins, denoise_fn=DIFF_DECODERS[hparams['diff_decoder_type']](hparams),
            timesteps=hparams['timesteps'],
            K_step=hparams['K_step'],
            loss_type=hparams['diff_loss_type'],
            spec_min=hparams['spec_min'], spec_max=hparams['spec_max'],
        )
        # if hparams['fs2_ckpt'] != '':
        #     utils.load_ckpt(self.model.fs2, hparams['fs2_ckpt'], 'model', strict=True)
        #     self.model.fs2.decoder = None

    def run_model(self, model, sample, return_output=False, infer=False):
        txt_tokens = sample['txt_tokens']  # [B, T_t]