Add configurable WORKERS env var, default to 1 for GPU models

CLAUDE.md

@@ -77,6 +77,11 @@ owned_by = "UncloseAI"
 - `sample.env` - Default environment (commit this)
 - `speech.env` - Runtime environment (created automatically by Makefile)
 
+**Key environment variables:**
+- `WORKERS` - Number of uvicorn workers (default: 1 for GPU models like Qwen3-TTS)
+  - Use 1 for GPU-bound models to avoid VRAM duplication across workers
+  - Increase for CPU-bound models like Piper (e.g., WORKERS=4)
+
 **Never view or log secrets** - source them and use them.
 
 ### 4. Git Commit Guidelines

sample.env

@@ -1,5 +1,10 @@
 TTS_HOME=voices
 HF_HOME=voices
+
+# Worker processes (default: 1 for GPU models like Qwen3-TTS)
+# Increase for CPU-bound models like Piper
+WORKERS=1
+
 #PRELOAD_MODEL=xtts
 #PRELOAD_MODEL=xtts_v2.0.2
 #EXTRA_ARGS=--log-level DEBUG --unload-timer 300

startup.sh

@@ -2,6 +2,9 @@
 
 [ -f speech.env ] && . speech.env
 
+# Default to 1 worker for GPU models (Qwen3-TTS)
+WORKERS=${WORKERS:-1}
+
 echo "First startup may download ~3GB of Qwen3-TTS model. Please wait."
 
 # Pre-download Qwen3-TTS model (default engine)
@@ -11,4 +14,4 @@ python -c "from qwen_tts import Qwen3TTSModel; Qwen3TTSModel.from_pretrained('Qw
 # bash download_voices_tts-1.sh
 # bash download_voices_tts-1-hd.sh $PRELOAD_MODEL
 
-python speech.py ${PRELOAD_MODEL:+--preload $PRELOAD_MODEL} $EXTRA_ARGS $@
+python speech.py --workers $WORKERS ${PRELOAD_MODEL:+--preload $PRELOAD_MODEL} $EXTRA_ARGS $@