remove speech enhancement

app.py

@@ -4,12 +4,10 @@ import torch
 
 from transformers import pipeline
 
-from resemble_enhance.enhancer.inference import denoise, enhance
-
 checkpoint_finetuned = "JackismyShephard/speecht5_tts-finetuned-nst-da"
 
 revision = "5af228df418092b681cf31c31e413bdd2b5f9c8c"
-device = 0 if torch.cuda.is_available() else "cpu"
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
 pipe = pipeline(
     "text-to-speech",
@@ -35,7 +33,7 @@ target_dtype = np.int16
 max_range = np.iinfo(target_dtype).max
 
 
-def predict(text, speaker, post_process):
+def predict(text, speaker):
     if len(text.strip()) == 0:
         return (16000, np.zeros(0))
 
@@ -52,10 +50,7 @@ def predict(text, speaker, post_process):
     forward_params = {"speaker_embeddings": speaker_embedding}
     speech = pipe(text, forward_params=forward_params)
 
-    if post_process:
-        sr, audio = enhance_audio(speech["audio"], speech["sampling_rate"], device)
-    else:
-        sr, audio = speech["sampling_rate"], speech["audio"]
+    sr, audio = speech["sampling_rate"], speech["audio"]
 
     audio = (audio * max_range).astype(np.int16)
 
@@ -90,17 +85,6 @@ replacements = [
     ("ü", "y"),
 ]
 
-
-def enhance_audio(waveform, sr, device="cuda"):
-    tensor = torch.tensor(waveform).float()
-    denoised, new_sr = denoise(tensor, sr, device)
-    enhanced, new_sr = enhance(
-        denoised, new_sr, device, nfe=64, solver="midpoint", lambd=0.1, tau=0.5
-    )
-    enhanced_cpu = enhanced.cpu().numpy()
-    return new_sr, enhanced_cpu
-
-
 title = "Danish Speech Synthesis"
 
 description = (
@@ -113,7 +97,6 @@ examples = [
     [
         "I sin oprindelige før-kristne form blev alferne sandsynligvis opfattet som en personificering af det land og den natur, der omgav menneskene, dvs. den opdyrkede jord, gården og de naturressourcer, som hørte dertil. De var guddommelige eller delvis guddommelige væsener, der besad magiske kræfter, som de brugte både til fordel og ulempe for menneskene.",
         "F23 (Female, 23, Vestjylland)",
-        True,
     ],
 ]
 
@@ -133,7 +116,6 @@ demo = gr.Interface(
             ],
             value="F23 (Female, 23, Vestjylland)",
         ),
-        gr.Checkbox(label="Enhance audio (takes substantially longer)"),
     ],
     outputs=[
         gr.Audio(label="Generated Speech", type="numpy"),

requirements.txt

@@ -1,4 +1,3 @@
 torch
 transformers
-sentencepiece
-resemble-enhance
+sentencepiece