Fix async iteration over vLLM generator - use async for instead of sync for

main.py

@@ -164,9 +164,9 @@ def get_custom_voices() -> List[str]:
     return voices
 
 
-def generate_speech(text: str, voice: str) -> bytes:
+async def generate_speech(text: str, voice: str) -> bytes:
     """
-    Generate speech using Orpheus model.
+    Generate speech using Orpheus model (async version).
     
     Args:
         text: Text to convert (may include emotion tags)
@@ -189,8 +189,10 @@ def generate_speech(text: str, voice: str) -> bytes:
         print(f"Unknown voice '{voice}', using default '{DEFAULT_VOICE}'")
         voice = DEFAULT_VOICE
     
-    # Generate speech using Orpheus
+    print(f"{text}")
-    # Note: text is passed as-is, emotion tags like <laugh> are handled by Orpheus
+    
+    # Generate speech using Orpheus - async iteration!
+    import numpy as np
     audio_chunks = []
     
     syn_tokens = model.generate_speech(
@@ -198,12 +200,11 @@ def generate_speech(text: str, voice: str) -> bytes:
         voice=voice,
     )
     
-    # Collect audio chunks
+    # Async iteration over the generator
-    for audio_chunk in syn_tokens:
+    async for audio_chunk in syn_tokens:
         audio_chunks.append(audio_chunk)
     
     # Combine chunks into single audio
-    import numpy as np
     audio_data = np.concatenate(audio_chunks) if len(audio_chunks) > 1 else audio_chunks[0]
     
     # Convert to WAV bytes
@@ -225,8 +226,8 @@ def save_audio_to_file(job_id: str, audio_bytes: bytes) -> str:
     return str(output_path)
 
 
-def generate_speech_background(job_id: str, text: str, voice: str):
+async def generate_speech_background(job_id: str, text: str, voice: str):
-    """Background task for speech generation."""
+    """Background task for speech generation (async)."""
     try:
         jobs[job_id].status = JobStatus.PROCESSING
         jobs[job_id].progress = 25
@@ -251,7 +252,7 @@ def generate_speech_background(job_id: str, text: str, voice: str):
         save_jobs_to_disk()
 
         print(f"Generating audio for job {job_id}...")
-        audio_bytes = generate_speech(text, voice)
+        audio_bytes = await generate_speech(text, voice)
 
         # Save to file
         jobs[job_id].progress = 75
@@ -437,7 +438,7 @@ def list_voices():
 
 
 @app.post("/tts/submit", response_model=JobResponse)
-async def submit_tts_job(request: TTSRequest, background_tasks: BackgroundTasks):
+async def submit_tts_job(request: TTSRequest):
     """Submit a TTS job for processing."""
     job_id = str(uuid.uuid4())
 
@@ -453,11 +454,9 @@ async def submit_tts_job(request: TTSRequest, background_tasks: BackgroundTasks)
     jobs[job_id] = job
     save_jobs_to_disk()
 
-    background_tasks.add_task(
+    # Use asyncio.create_task for proper async execution
-        generate_speech_background,
+    asyncio.create_task(
-        job_id,
+        generate_speech_background(job_id, request.text, request.voice)
-        request.text,
-        request.voice
     )
 
     print(f"Job {job_id} submitted: '{request.text[:50]}...' with voice '{request.voice}'")
@@ -530,14 +529,14 @@ async def stream_tts(request: TTSStreamRequest):
         voice = DEFAULT_VOICE
     
     async def audio_generator():
-        """Generate audio chunks"""
+        """Generate audio chunks (async)"""
         try:
             syn_tokens = model.generate_speech(
                 prompt=request.text,
                 voice=voice,
             )
             
-            for audio_chunk in syn_tokens:
+            async for audio_chunk in syn_tokens:
                 yield audio_chunk
                 
         except Exception as e: