fixed regression where the auto_conds do not move to the GPU and causes a problem during CVVP compare pass

Reviewed-on: mrq/tortoise-tts#20

tortoise/api.py

@@ -483,8 +483,11 @@ class TextToSpeech:
                 auto_conditioning, diffusion_conditioning, auto_conds, _ = conditioning_latents
         else:
             auto_conditioning, diffusion_conditioning = self.get_random_conditioning_latents()
+
         auto_conditioning = auto_conditioning.to(self.device)
         diffusion_conditioning = diffusion_conditioning.to(self.device)
+        if auto_conds is not None:
+            auto_conds = auto_conds.to(self.device)
 
         diffuser = load_discrete_vocoder_diffuser(desired_diffusion_steps=diffusion_iterations, cond_free=cond_free, cond_free_k=cond_free_k)
 
@@ -539,8 +542,10 @@ class TextToSpeech:
                 for batch in tqdm_override(samples, verbose=verbose, progress=progress, desc=desc):
                     for i in range(batch.shape[0]):
                         batch[i] = fix_autoregressive_output(batch[i], stop_mel_token)
+
                     if cvvp_amount != 1:
                         clvp = self.clvp(text_tokens.repeat(batch.shape[0], 1), batch, return_loss=False)
+                        
                     if auto_conds is not None and cvvp_amount > 0:
                         cvvp_accumulator = 0
                         for cl in range(auto_conds.shape[1]):

webui.py

@@ -265,7 +265,7 @@ def generate(
         with open(f'{get_voice_dir()}/{voice}/cond_latents.pth', 'rb') as f:
             info['latents'] = base64.b64encode(f.read()).decode("ascii")
 
-    if voicefixer:
+    if args.voice_fixer and voicefixer:
         # we could do this on the pieces before they get stiched up anyways to save some compute
         # but the stitching would need to read back from disk, defeating the point of caching the waveform
         for path in progress.tqdm(audio_cache, desc="Running voicefix..."):
@@ -443,8 +443,12 @@ def reload_tts():
 def cancel_generate():
     tortoise.api.STOP_SIGNAL = True
 
+def get_voice_list():
+    voice_dir = get_voice_dir()
+    return [d for d in os.listdir(voice_dir) if os.path.isdir(os.path.join(voice_dir, d))]
+
 def update_voices():
-    return gr.Dropdown.update(choices=sorted(os.listdir(get_voice_dir())) + ["microphone"])
+    return gr.Dropdown.update(choices=sorted(get_voice_list()) + ["microphone"])
 
 def export_exec_settings( share, listen, check_for_updates, models_from_local_only, low_vram, embed_output_metadata, latents_lean_and_mean, voice_fixer, cond_latent_max_chunk_size, sample_batch_size, concurrency_count, output_sample_rate, output_volume ):
     args.share = share
@@ -592,7 +596,7 @@ def setup_gradio():
                     )
                     prompt = gr.Textbox(lines=1, label="Custom Emotion + Prompt (if selected)")
                     voice = gr.Dropdown(
-                        sorted(os.listdir(get_voice_dir())) + ["microphone"],
+                        sorted(get_voice_list()) + ["microphone"],
                         label="Voice",
                         type="value",
                     )
@@ -692,7 +696,7 @@ def setup_gradio():
             with gr.Row():
                 with gr.Column():
                     history_voices = gr.Dropdown(
-                        sorted(os.listdir(get_voice_dir())) + ["microphone"],
+                        sorted(get_voice_list()) + ["microphone"],
                         label="Voice",
                         type="value",
                     )