more work on streaming timestamps (not working weird error) :(

Test.py

@@ -12,10 +12,7 @@ response = requests.post(
         "input": "http://localhost:8880/web/",
         "voice": "af_heart",
         "response_format": "mp3",  # Supported: mp3, wav, opus, flac
-        "speed": 1.0,
+        "speed": 1.0
-        "normalization_options": {
-            "normalize": True
-        }
     }
 )
 

api/src/inference/kokoro_v1.py

@@ -291,17 +291,6 @@ class KokoroV1(BaseModelBackend):
                                         f"Added timestamp for word '{token.text}': {start_time:.3f}s - {end_time:.3f}s"
                                     )
 
-                                # Update offset for next chunk based on pred_dur
-                                chunk_duration = (
-                                    float(result.pred_dur.sum()) / 80
-                                )  # Convert frames to seconds
-                                current_offset = max(
-                                    current_offset + chunk_duration, end_time
-                                )
-                                logger.debug(
-                                    f"Updated time offset to {current_offset:.3f}s"
-                                )
-
                             except Exception as e:
                                 logger.error(
                                     f"Failed to process timestamps for chunk: {e}"

api/src/routers/openai_compatible.py

@@ -8,7 +8,7 @@ import tempfile
 from typing import AsyncGenerator, Dict, List, Union, Tuple
 
 import aiofiles
-from inference.base import AudioChunk
+from ..inference.base import AudioChunk
 import torch
 from fastapi import APIRouter, Depends, Header, HTTPException, Request, Response
 from fastapi.responses import FileResponse, StreamingResponse
@@ -214,16 +214,16 @@ async def create_speech(
                 }
 
                 # Create async generator for streaming
-                async def dual_output(return_json:bool=False):
+                async def dual_output():
                     try:
                         # Write chunks to temp file and stream
                         async for chunk, chunk_data in generator:
                             if chunk:  # Skip empty chunks
                                 await temp_writer.write(chunk)
-                                if return_json:
+                                #if return_json:
-                                    yield chunk, chunk_data
+                                #    yield chunk, chunk_data
-                                else:
+                                #else:
-                                    yield chunk
+                                yield chunk
 
                         # Finalize the temp file
                         await temp_writer.finalize()

api/src/services/tts_service.py

@@ -327,11 +327,22 @@ class TTSService:
         lang_code: Optional[str] = None,
     ) -> Tuple[Tuple[np.ndarray,AudioChunk]]:
         """Generate complete audio for text using streaming internally."""
+        start_time = time.time()
+        audio_chunks = []
+        audio_data_chunks=[]
+        word_timestamps = []
+        
         start_time = time.time()
         chunks = []
         word_timestamps = []
-
         try:
+            async for audio_stream,audio_stream_data in self.generate_audio_stream(text,voice,speed=speed,return_timestamps=return_timestamps,lang_code=lang_code):
+                print("common")
+                audio_chunks.append(audio_stream_data.audio)
+                audio_data_chunks.append(audio_stream_data)
+                
+            print(audio_data_chunks)
+            """
             # Get backend and voice path
             backend = self.model_manager.get_backend()
             voice_name, voice_path = await self._get_voice_path(voice)
@@ -574,10 +585,11 @@ class TTSService:
                         [],
                     )  # Empty timestamps for legacy backends
                 return audio, processing_time
-
+        """
         except Exception as e:
             logger.error(f"Error in audio generation: {str(e)}")
             raise
+        
 
     async def combine_voices(self, voices: List[str]) -> torch.Tensor:
         """Combine multiple voices.