Update voice_processing.py

voice_processing.py

@@ -34,7 +34,7 @@ limitation = os.getenv("SYSTEM") == "spaces"
 
 config = Config()
 
-# Edge TTS
+# Edge TTS voices
 tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices())
 tts_voices = ["mn-MN-BataaNeural", "mn-MN-YesuiNeural"]  # Specific voices
 
@@ -47,12 +47,13 @@ def get_unique_filename(extension):
     return f"{uuid.uuid4()}.{extension}"
 
 def model_data(model_name):
-    pth_path = [
-        f"{model_root}/{model_name}/{f}"
-        for f in os.listdir(f"{model_root}/{model_name}")
-        if f.endswith(".pth")
-    ][0]
-    print(f"Loading {pth_path}")
+    pth_files = [
+        f for f in os.listdir(f"{model_root}/{model_name}") if f.endswith(".pth")
+    ]
+    if not pth_files:
+        raise FileNotFoundError(f"No .pth file found for model '{model_name}'")
+    pth_path = f"{model_root}/{model_name}/{pth_files[0]}"
+    print(f"Loading model from {pth_path}")
     cpt = torch.load(pth_path, map_location="cpu")
     tgt_sr = cpt["config"][-1]
     cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]  # n_spk
@@ -72,25 +73,23 @@ def model_data(model_name):
         raise ValueError("Unknown version")
     del net_g.enc_q
     net_g.load_state_dict(cpt["weight"], strict=False)
-    print("Model loaded")
     net_g.eval().to(config.device)
     if config.is_half:
         net_g = net_g.half()
     else:
         net_g = net_g.float()
+    print(f"Model '{model_name}' loaded.")
     vc = VC(tgt_sr, config)
 
     index_files = [
-        f"{model_root}/{model_name}/{f}"
-        for f in os.listdir(f"{model_root}/{model_name}")
-        if f.endswith(".index")
+        f for f in os.listdir(f"{model_root}/{model_name}") if f.endswith(".index")
     ]
-    if len(index_files) == 0:
-        print("No index file found")
-        index_file = ""
-    else:
-        index_file = index_files[0]
+    if index_files:
+        index_file = f"{model_root}/{model_name}/{index_files[0]}"
         print(f"Index file found: {index_file}")
+    else:
+        index_file = ""
+        print("No index file found.")
 
     return tgt_sr, net_g, vc, version, index_file, if_f0
 
@@ -108,22 +107,17 @@ def load_hubert():
     return hubert_model.eval()
 
 def get_model_names():
-    model_root = "weights"  # Assuming this is where your models are stored
     return [d for d in os.listdir(model_root) if os.path.isdir(f"{model_root}/{d}")]
 
-# Add this helper function to ensure a new event loop is created if none exists
-def run_async_in_thread(fn, *args):
-    loop = asyncio.new_event_loop()
-    asyncio.set_event_loop(loop)
-    result = loop.run_until_complete(fn(*args))
-    loop.close()
-    return result
+# Voice mapping dictionary
+voice_mapping = {
+    "Mongolian Male": "mn-MN-BataaNeural",
+    "Mongolian Female": "mn-MN-YesuiNeural"
+}
 
-def parallel_tts(tasks):
-    with ThreadPoolExecutor() as executor:
-        futures = [executor.submit(run_async_in_thread, tts, *task) for task in tasks]
-        results = [future.result() for future in futures]
-    return results
+# Load models once
+hubert_model = load_hubert()
+rmvpe_model = RMVPE("rmvpe.pt", config.is_half, config.device)
 
 async def tts(
     model_name,
@@ -133,63 +127,65 @@ async def tts(
     use_uploaded_voice,
     uploaded_voice,
 ):
-    # Default values for parameters used in EdgeTTS
-    speed = 0  # Default speech speed
-    f0_up_key = 0  # Default pitch adjustment
-    f0_method = "rmvpe"  # Default pitch extraction method
-    protect = 0.33  # Default protect value
-    filter_radius = 3
-    resample_sr = 0
-    rms_mix_rate = 0.25
-    edge_time = 0  # Initialize edge_time
-
-    edge_output_filename = get_unique_filename("mp3")
-
     try:
+        # Validate inputs
+        if not tts_text.strip():
+            return {"success": False, "error": "Input text is empty."}
+
+        if tts_voice not in voice_mapping.values():
+            return {"success": False, "error": f"Invalid voice '{tts_voice}'."}
+
+        # Default parameters
+        f0_up_key = 0  # Pitch adjustment
+        f0_method = "rmvpe"  # Pitch extraction method
+        protect = 0.33  # Protect value
+        filter_radius = 3
+        resample_sr = 0
+        rms_mix_rate = 0.25
+        edge_time = 0
+
+        audio = None
+        sr = 16000  # Sample rate
+
         if use_uploaded_voice:
             if uploaded_voice is None:
-                return "No voice file uploaded.", None, None
-            
+                return {"success": False, "error": "No voice file uploaded."}
+
             # Process the uploaded voice file
             with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
                 tmp_file.write(uploaded_voice)
-                uploaded_file_path = tmp_file.name
+                input_audio_path = tmp_file.name
 
-            audio, sr = librosa.load(uploaded_file_path, sr=16000, mono=True)
+            audio, sr = librosa.load(input_audio_path, sr=16000, mono=True)
         else:
-            # EdgeTTS processing
+            # Edge TTS processing
+            edge_output_filename = get_unique_filename("mp3")
+
+            # Edge TTS limitations
             if limitation and len(tts_text) > 12000:
-                return (
-                    f"Text characters should be at most 12000 in this huggingface space, but got {len(tts_text)} characters.",
-                    None,
-                    None,
-                )
-            
-            # Invoke Edge TTS
-            t0 = time.time()
+                return {
+                    "success": False,
+                    "error": f"Text characters should be at most 12000 in this huggingface space, but got {len(tts_text)} characters."
+                }
+
+            speed = 0  # Speech speed
             speed_str = f"+{speed}%" if speed >= 0 else f"{speed}%"
-            await edge_tts.Communicate(
+            communicate = edge_tts.Communicate(
                 tts_text, tts_voice, rate=speed_str
-            ).save(edge_output_filename)
+            )
+
+            t0 = time.time()
+            await communicate.save(edge_output_filename)
             t1 = time.time()
             edge_time = t1 - t0
 
             audio, sr = librosa.load(edge_output_filename, sr=16000, mono=True)
+            input_audio_path = edge_output_filename
 
-        # Common processing after loading the audio
-        duration = len(audio) / sr
-        print(f"Audio duration: {duration}s")
-        if limitation and duration >= 20000:
-            return (
-                f"Audio should be less than 20 seconds in this huggingface space, but got {duration}s.",
-                None,
-                None,
-            )
-
-        f0_up_key = int(f0_up_key)
+        # Load the specified RVC model
         tgt_sr, net_g, vc, version, index_file, if_f0 = model_data(model_name)
 
-        # Setup for RMVPE or other pitch extraction methods
+        # Set RMVPE model for pitch extraction
         if f0_method == "rmvpe":
             vc.model_rmvpe = rmvpe_model
 
@@ -198,9 +194,9 @@ async def tts(
         audio_opt = vc.pipeline(
             hubert_model,
             net_g,
-            0,
+            0,  # Speaker ID
             audio,
-            edge_output_filename if not use_uploaded_voice else uploaded_file_path,
+            input_audio_path,
             times,
             f0_up_key,
             f0_method,
@@ -218,25 +214,25 @@ async def tts(
 
         if tgt_sr != resample_sr and resample_sr >= 16000:
             tgt_sr = resample_sr
-        
-        info = f"Success. Time: tts: {edge_time}s, npy: {times[0]}s, f0: {times[1]}s, infer: {times[2]}s"
+
+        info = f"Success. Time: tts: {edge_time:.2f}s, npy: {times[0]:.2f}s, f0: {times[1]:.2f}s, infer: {times[2]:.2f}s"
         print(info)
-        return (
-            info,
-            edge_output_filename if not use_uploaded_voice else None,   
-            (tgt_sr, audio_opt),
-        )
+        return {
+            "success": True,
+            "info": info,
+            "tgt_sr": tgt_sr,
+            "audio_opt": audio_opt
+        }
 
     except EOFError:
-        info = (
-            "output not valid. This may occur when input text and speaker do not match."
-        )
+        info = "Output not valid. This may occur when input text and speaker do not match."
         print(info)
-        return info, None, None
+        return {"success": False, "error": info}
     except Exception as e:
         traceback_info = traceback.format_exc()
         print(traceback_info)
-        return str(e), None, None
+        return {"success": False, "error": str(e)}
+
 
 voice_mapping = {
     "Mongolian Male": "mn-MN-BataaNeural",