Simplifed generate_audio in tts_service mostly working (audio conversion does not work)

Test.py

@@ -12,7 +12,8 @@ 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,
+        "stream":False,
     }
 )
 

api/src/inference/base.py

@@ -16,6 +16,17 @@ class AudioChunk:
         self.audio=audio
         self.word_timestamps=word_timestamps
         
+    @staticmethod
+    def combine(audio_chunk_list: List):
+        output=AudioChunk(audio_chunk_list[0].audio,audio_chunk_list[0].word_timestamps)
+        
+        for audio_chunk in audio_chunk_list[1:]:
+            output.audio=np.concatenate((output.audio,audio_chunk.audio))
+            if output.word_timestamps is not None:
+                output.word_timestamps+=output.word_timestamps
+                
+        return output
+            
 class ModelBackend(ABC):
     """Abstract base class for model inference backend."""
 

api/src/routers/openai_compatible.py

@@ -6,6 +6,7 @@ import os
 import re
 import tempfile
 from typing import AsyncGenerator, Dict, List, Union, Tuple
+from urllib import response
 
 import aiofiles
 from ..inference.base import AudioChunk
@@ -141,7 +142,7 @@ async def stream_audio_chunks(
             output_format=request.response_format,
             lang_code=request.lang_code or request.voice[0],
             normalization_options=request.normalization_options,
-            return_timestamps=True,
+            return_timestamps=False,
         ):
 
             # Check if client is still connected
@@ -258,7 +259,7 @@ async def create_speech(
             )
         else:
             # Generate complete audio using public interface
-            audio, _ = await tts_service.generate_audio(
+            audio, audio_data = await tts_service.generate_audio(
                 text=request.input,
                 voice=voice_name,
                 speed=request.speed,
@@ -266,14 +267,14 @@ async def create_speech(
             )
 
             # Convert to requested format with proper finalization
-            content = await AudioService.convert_audio(
-                audio,
+            content, audio_data = await AudioService.convert_audio(
+                audio_data,
                 24000,
                 request.response_format,
                 is_first_chunk=True,
                 is_last_chunk=True,
             )
-
+            print(content,request.response_format)
             return Response(
                 content=content,
                 media_type=content_type,

api/src/services/audio.py

@@ -148,9 +148,11 @@ class AudioService:
             if normalizer is None:
                 normalizer = AudioNormalizer()
             
+            print(len(audio_chunk.audio),"1")
             audio_chunk.audio = await normalizer.normalize(audio_chunk.audio)
+            print(len(audio_chunk.audio),"2")
             audio_chunk = AudioService.trim_audio(audio_chunk,chunk_text,speed,is_last_chunk,normalizer)
-
+            print(len(audio_chunk.audio),"3")
             # Get or create format-specific writer
             writer_key = f"{output_format}_{sample_rate}"
             if is_first_chunk or writer_key not in AudioService._writers:
@@ -167,6 +169,7 @@ class AudioService:
             if is_last_chunk:
                 final_data = writer.write_chunk(finalize=True)
                 del AudioService._writers[writer_key]
+                print(audio_chunk.audio)
                 return final_data if final_data else b"", audio_chunk
 
             return chunk_data if chunk_data else b"", audio_chunk

api/src/services/tts_service.py

@@ -333,25 +333,27 @@ class TTSService:
         text: str,
         voice: str,
         speed: float = 1.0,
-        return_timestamps: bool = False,
+        return_timestamps: bool = True,
         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)
+            print(audio_data_chunks[0].audio.shape)
+            
+            combined_audio=np.concatenate(audio_chunks)
+            print("1")
+            combined_audio_data=AudioChunk.combine(audio_data_chunks)
+            print("2")
+            print(len(combined_audio_data.audio))
+            return combined_audio,combined_audio_data
             """
             # Get backend and voice path
             backend = self.model_manager.get_backend()