3b57e88a66ba1f410be70dbd2ad2899b5b1bcb0e,server/synthesizer.py,Synthesizer,tts,#Synthesizer#Any#,155

Before Change

                postnet_output, decoder_output, alignments)

            if self.wavernn:
                postnet_output = postnet_output[0].data.cpu().numpy()
                wav = self.wavernn.generate(torch.FloatTensor(postnet_output.T).unsqueeze(0).cuda(), batched=self.config.is_wavernn_batched, target=11000, overlap=550)
            else:
                wav = inv_spectrogram(postnet_output, self.ap, self.tts_config)
            // trim silence

After Change

                    vocoder_input.cuda()
                wav = self.pwgan.inference(vocoder_input, hop_size=self.ap.hop_length)
            if self.wavernn:
                vocoder_input = torch.FloatTensor(postnet_output.T).unsqueeze(0)
                if self.use_cuda:
                    vocoder_input.cuda()
                wav = self.wavernn.generate(vocoder_input, batched=self.config.is_wavernn_batched, target=11000, overlap=550)
            else:
                wav = inv_spectrogram(postnet_output, self.ap, self.tts_config)
            // trim silence
            wav = trim_silence(wav, self.ap)

Instances