fix: Play produces no sound on HF Space

Root causes:
- Stock voice profile was only created lazily on first Play, failing silently
- synthesize_custom_voice_streaming returned empty when profile missing
- Worker put empty arrays into queue, Gradio received zero-length audio

Fixes:
- Pre-create stock voice profile at startup (after model load)
- Raise RuntimeError instead of silent return when profile unavailable
- _ensure_stock_voice_profile raises FileNotFoundError if ref wav missing
- Skip empty wav segments in audio stream generator
- Guard against None wav in error path

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

app.py

@@ -410,10 +410,15 @@ else:
 
 # Preload models at startup for fast response
 print("[MomsVoice] Preloading Qwen3-TTS 1.7B Base model (cloning + stock voice)...")
-from voice_clone import get_qwen_tts
+from voice_clone import get_qwen_tts, _ensure_stock_voice_profile
 get_qwen_tts()
 print("[MomsVoice] TTS model ready.")
 
+# Pre-create stock voice profile so Play works immediately
+print("[MomsVoice] Creating stock voice profile (vivian)...")
+_ensure_stock_voice_profile()
+print("[MomsVoice] Stock voice profile ready.")
+
 print("[MomsVoice] Preloading LFM2.5-Audio-1.5B Q&A model...")
 from inference_lfm import get_lfm_model
 get_lfm_model()

tts.py

@@ -405,9 +405,15 @@ def generate_audio_stream(
             if audio_buffer:
                 combined = np.concatenate(audio_buffer)
                 yield sample_rate, combined, last_idx, n, None
-            yield sample_rate, wav, i, n, err
+                audio_buffer = []
+                buffer_samples = 0
+            yield sample_rate, np.zeros(0, dtype=np.float32), i, n, err
             break
         last_idx = i
+
+        # Skip empty audio segments
+        if wav is None or len(wav) == 0:
+            continue
         
         # Add transition chime between chunks (not before first)
         if add_transitions and chunk_count > 0:

voice_clone.py

@@ -373,12 +373,11 @@ def _ensure_stock_voice_profile():
         if STOCK_VOICE_PROFILE_ID in _PROFILE_CACHE:
             return
     # Create profile from vivian reference audio
-    if VIVIAN_REF_PATH.exists():
-        try:
-            create_voice_profile(str(VIVIAN_REF_PATH), voice_name="Vivian (Stock)", profile_id_override=STOCK_VOICE_PROFILE_ID)
-            logger.info("Stock vivian voice profile created from reference audio.")
-        except Exception as e:
-            logger.warning("Failed to create stock voice profile: %s", e)
+    if not VIVIAN_REF_PATH.exists():
+        logger.error("Stock voice reference not found at %s", VIVIAN_REF_PATH)
+        raise FileNotFoundError(f"Stock voice reference not found: {VIVIAN_REF_PATH}")
+    create_voice_profile(str(VIVIAN_REF_PATH), voice_name="Vivian (Stock)", profile_id_override=STOCK_VOICE_PROFILE_ID)
+    logger.info("Stock vivian voice profile created from reference audio.")
 
 
 def synthesize_custom_voice_streaming(
@@ -392,8 +391,10 @@ def synthesize_custom_voice_streaming(
     with _cache_lock:
         prompt_items = _PROFILE_CACHE.get(STOCK_VOICE_PROFILE_ID)
     if prompt_items is None:
-        logger.warning("Stock voice profile not available, synthesizing without clone.")
-        return
+        raise RuntimeError(
+            "Stock voice profile could not be created. "
+            "Check that assets/vivian_reference.wav exists and the TTS model loaded correctly."
+        )
 
     model = get_qwen_tts()
     with GPU_INFERENCE_LOCK: