Removed some unused code

synthesizer/synthesize.py

@@ -82,29 +82,3 @@ def run_synthesis(in_dir, out_dir, model_dir, hparams):
     print("Synthesized mel spectrograms at {}".format(synth_dir))
     return meta_out_fpath
 
-
-def tacotron_synthesize(args, hparams, checkpoint, sentences=None):
-    output_dir = args.output_dir
-    
-    try:
-        checkpoint_path = tf.train.get_checkpoint_state(checkpoint).model_checkpoint_path
-        log("loaded model at {}".format(checkpoint_path))
-    except:
-        raise RuntimeError("Failed to load checkpoint at {}".format(checkpoint))
-    
-    if hparams.tacotron_synthesis_batch_size < hparams.tacotron_num_gpus:
-        raise ValueError("Defined synthesis batch size {} is smaller than minimum required {} "
-                         "(num_gpus)! Please verify your synthesis batch size choice.".format(
-            hparams.tacotron_synthesis_batch_size, hparams.tacotron_num_gpus))
-    
-    if hparams.tacotron_synthesis_batch_size % hparams.tacotron_num_gpus != 0:
-        raise ValueError("Defined synthesis batch size {} is not a multiple of {} (num_gpus)! "
-                         "Please verify your synthesis batch size choice!".format(
-            hparams.tacotron_synthesis_batch_size, hparams.tacotron_num_gpus))
-    
-    if args.mode == "eval":
-        return run_eval(args, checkpoint_path, output_dir, hparams, sentences)
-    elif args.mode == "synthesis":
-        return run_synthesis(args, checkpoint_path, output_dir, hparams)
-    else:
-        run_live(args, checkpoint_path, hparams)

synthesizer/train.py

@@ -330,46 +330,13 @@ def train(log_dir, args, hparams):
                     saver.save(sess, checkpoint_fpath, global_step=global_step)
                     
                     log("\nSaving alignment, Mel-Spectrograms and griffin-lim inverted waveform..")
-                    if hparams.predict_linear:
-                        input_seq, mel_prediction, linear_prediction, alignment, target, target_length, linear_target = sess.run(
-                            [
-                                model.tower_inputs[0][0],
-                                model.tower_mel_outputs[0][0],
-                                model.tower_linear_outputs[0][0],
-                                model.tower_alignments[0][0],
-                                model.tower_mel_targets[0][0],
-                                model.tower_targets_lengths[0][0],
-                                model.tower_linear_targets[0][0],
-                            ])
-                        
-                        # save predicted linear spectrogram to disk (debug)
-                        linear_filename = "linear-prediction-step-{}.npy".format(step)
-                        np.save(os.path.join(linear_dir, linear_filename), linear_prediction.T,
-                                allow_pickle=False)
-                        
-                        # save griffin lim inverted wav for debug (linear -> wav)
-                        wav = audio.inv_linear_spectrogram(linear_prediction.T, hparams)
-                        audio.save_wav(wav, os.path.join(wav_dir, "step-{}-wave-from-linear.wav"
-                                                                  "".format(step)),
-                                       sr=hparams.sample_rate)
-                        
-                        # Save real and predicted linear-spectrogram plot to disk (control purposes)
-                        plot.plot_spectrogram(linear_prediction, os.path.join(plot_dir,
-                                                                              "step-{}-linear-spectrogram.png".format(
-                                                                                  step)),
-                                              title="{}, {}, step={}, loss={:.5f}".format(
-                                                  "Tacotron", time_string(), step, loss),
-                                              target_spectrogram=linear_target,
-                                              max_len=target_length, auto_aspect=True)
-                    
-                    else:
-                        input_seq, mel_prediction, alignment, target, target_length = sess.run([
-                            model.tower_inputs[0][0],
-                            model.tower_mel_outputs[0][0],
-                            model.tower_alignments[0][0],
-                            model.tower_mel_targets[0][0],
-                            model.tower_targets_lengths[0][0],
-                        ])
+                    input_seq, mel_prediction, alignment, target, target_length = sess.run([
+                        model.tower_inputs[0][0],
+                        model.tower_mel_outputs[0][0],
+                        model.tower_alignments[0][0],
+                        model.tower_mel_targets[0][0],
+                        model.tower_targets_lengths[0][0],
+                    ])
                     
                     # save predicted mel spectrogram to disk (debug)
                     mel_filename = "mel-prediction-step-{}.npy".format(step)

vocoder/display.py

@@ -13,7 +13,7 @@ def progbar(i, n, size=16):
 
 
 def stream(message) :
-    sys.stdout.write(f"\r{message}")
+    sys.stdout.write("\r{%s}" % message)
 
 
 def simple_table(item_tuples) :

vocoder/models/deepmind_version.py

@@ -4,6 +4,7 @@ import torch.nn.functional as F
 from utils.display import *
 from utils.dsp import *
 
+
 class WaveRNN(nn.Module) :
     def __init__(self, hidden_size=896, quantisation=256) :
         super(WaveRNN, self).__init__()