Update orpheus-tts/engine_class.py

orpheus-tts/engine_class.py

@@ -6,6 +6,7 @@ from transformers import AutoTokenizer
 import threading
 import queue
 from decoder import tokens_decoder_sync
+from kartoffel_decoder import tokens_decoder_kartoffel_sync
 
 class OrpheusModel:
     def __init__(self, model_name, dtype=torch.bfloat16, tokenizer=None, **engine_kwargs):
@@ -86,7 +87,7 @@ class OrpheusModel:
             if voice not in self.engine.available_voices:
                 raise ValueError(f"Voice {voice} is not available for model {self.model_name}")
     
-    def _format_prompt(self, prompt, voice="Sophie", model_type="larger"):
+    def _format_prompt(self, prompt, voice="Jakob", model_type="larger"):
         # Use Kartoffel model format based on documentation
         if voice:
             full_prompt = f"{voice}: {prompt}"
@@ -166,9 +167,15 @@ class OrpheusModel:
             token_generator = self.generate_tokens_sync(**kwargs)
             print("DEBUG: Token generator created successfully")
             
-            audio_generator = tokens_decoder_sync(token_generator)
-            print("DEBUG: Audio decoder called successfully")
+            # Verwende Kartoffel-Decoder für deutsche Modelle
+            if "german" in self.model_name.lower() or "kartoffel" in self.model_name.lower():
+                print("DEBUG: Using Kartoffel decoder for German model")
+                audio_generator = tokens_decoder_kartoffel_sync(token_generator, self.tokenizer)
+            else:
+                print("DEBUG: Using original decoder")
+                audio_generator = tokens_decoder_sync(token_generator)
             
+            print("DEBUG: Audio decoder called successfully")
             return audio_generator
         except Exception as e:
             print(f"DEBUG: Error in generate_speech: {e}")