updates to scripts

api.py

@@ -202,7 +202,7 @@ class TextToSpeech:
             'ultra_fast': {'num_autoregressive_samples': 32, 'diffusion_iterations': 16, 'cond_free': False},
             'fast': {'num_autoregressive_samples': 96, 'diffusion_iterations': 32},
             'standard': {'num_autoregressive_samples': 256, 'diffusion_iterations': 128},
-            'high_quality': {'num_autoregressive_samples': 512, 'diffusion_iterations': 2048},
+            'high_quality': {'num_autoregressive_samples': 512, 'diffusion_iterations': 1024},
         }
         kwargs.update(presets[preset])
         return self.tts(text, voice_samples, **kwargs)

do_tts.py

@@ -11,6 +11,10 @@ if __name__ == '__main__':
     parser.add_argument('--text', type=str, help='Text to speak.', default="I am a language model that has learned to speak.")
     parser.add_argument('--voice', type=str, help='Selects the voice to use for generation. See options in voices/ directory (and add your own!) '
                                                  'Use the & character to join two voices together. Use a comma to perform inference on multiple voices.', default='patrick_stewart')
+    parser.add_argument('--preset', type=str, help='Which voice preset to use.', default='standard')
+    parser.add_argument('--voice_diversity_intelligibility_slider', type=float,
+                        help='How to balance vocal diversity with the quality/intelligibility of the spoken text. 0 means highly diverse voice (not recommended), 1 means maximize intellibility',
+                        default=.5)
     parser.add_argument('--output_path', type=str, help='Where to store outputs.', default='results/')
     args = parser.parse_args()
     os.makedirs(args.output_path, exist_ok=True)
@@ -25,6 +29,6 @@ if __name__ == '__main__':
         for cond_path in cond_paths:
             c = load_audio(cond_path, 22050)
             conds.append(c)
-        gen = tts.tts_with_preset(args.text, conds, preset='standard')
+        gen = tts.tts_with_preset(args.text, conds, preset=args.preset, clvp_cvvp_slider=args.voice_diversity_intelligibility_slider)
         torchaudio.save(os.path.join(args.output_path, f'{voice}.wav'), gen.squeeze(0).cpu(), 24000)
 

read.py

@@ -28,11 +28,14 @@ def split_and_recombine_text(texts, desired_length=200, max_len=300):
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
-    parser.add_argument('--textfile', type=str, help='A file containing the text to read.', default="data/riding_hood2.txt")
+    parser.add_argument('--textfile', type=str, help='A file containing the text to read.', default="data/riding_hood.txt")
     parser.add_argument('--voice', type=str, help='Selects the voice to use for generation. See options in voices/ directory (and add your own!) '
                                                  'Use the & character to join two voices together. Use a comma to perform inference on multiple voices.', default='patrick_stewart')
     parser.add_argument('--output_path', type=str, help='Where to store outputs.', default='results/longform/')
-    parser.add_argument('--generation_preset', type=str, help='Preset to use for generation', default='standard')
+    parser.add_argument('--preset', type=str, help='Which voice preset to use.', default='standard')
+    parser.add_argument('--voice_diversity_intelligibility_slider', type=float,
+                        help='How to balance vocal diversity with the quality/intelligibility of the spoken text. 0 means highly diverse voice (not recommended), 1 means maximize intellibility',
+                        default=.5)
     args = parser.parse_args()
 
     outpath = args.output_path
@@ -60,16 +63,11 @@ if __name__ == '__main__':
         if not cond_paths:
             print('Error: no valid voices specified. Try again.')
 
-        priors = []
+        conds = []
+        for cond_path in cond_paths:
+            c = load_audio(cond_path, 22050)
+            conds.append(c)
         for j, text in enumerate(texts):
-            conds = priors.copy()
-            for cond_path in cond_paths:
-                c = load_audio(cond_path, 22050)
-                conds.append(c)
-            gen = tts.tts_with_preset(text, conds, preset=args.generation_preset)
+            gen = tts.tts_with_preset(text, conds, preset=args.preset, clvp_cvvp_slider=args.voice_diversity_intelligibility_slider)
             torchaudio.save(os.path.join(voice_outpath, f'{j}.wav'), gen.squeeze(0).cpu(), 24000)
 
-            priors.append(torchaudio.functional.resample(gen, 24000, 22050).squeeze(0))
-            while len(priors) > 2:
-                priors.pop(0)
-