diff --git a/Makefile b/Makefile
index 76cd55a..1ad53b4 100644
--- a/Makefile
+++ b/Makefile
@@ -14,7 +14,7 @@ REMOTE_USER ?= $(USER)
 REMOTE_PATH ?= ~/uncloseai-speech
 CONTAINER_NAME ?= uncloseai-speech-server-1
 
-.PHONY: help deploy sync restart logs test clean stop start voices voices-piper voices-xtts voices-kokoro test-kokoro voices-silero test-silero voices-chatterbox test-chatterbox push-all
+.PHONY: help deploy sync restart logs test clean stop start voices voices-piper voices-xtts voices-kokoro test-kokoro voices-silero test-silero voices-chatterbox test-chatterbox push-all hydrate load-test
 
 help:
 	@echo "🦝 Raccoon TTS Mission - Development Commands"
@@ -38,6 +38,10 @@ help:
 	@echo "  make voices-chatterbox - Download Chatterbox models"
 	@echo "  make test-chatterbox   - Test Chatterbox TTS with emotion control"
 	@echo ""
+	@echo "Testing:"
+	@echo "  make hydrate     - Hydrate all models by testing ALL 227 voices"
+	@echo "  make load-test   - Load test with concurrent random voice requests"
+	@echo ""
 	@echo "Container:"
 	@echo "  make start       - Start Docker container"
 	@echo "  make stop        - Stop Docker container"
@@ -209,3 +213,88 @@ test-chatterbox:
 		-o /tmp/chatterbox_test.mp3
 	@echo "✅ Test complete! Playing audio..."
 	@firefox /tmp/chatterbox_test.mp3 || mpv /tmp/chatterbox_test.mp3 || echo "Install firefox or mpv to play audio"
+
+hydrate:
+	@echo "🦝 Hydrating all TTS models by testing ALL voices..."
+	@echo "This will test all 227 voices across 4 engines (Piper, XTTS, Silero, Kokoro)"
+	@echo ""
+	@mkdir -p /tmp/hydrate_test
+	@curl -s http://$(REMOTE_HOST):8000/v1/voices | jq -r '.data[] as $$model | $$model.voices[] | "\($$model.id):\(.)"' > /tmp/hydrate_voices.txt
+	@TOTAL=$$(wc -l < /tmp/hydrate_voices.txt); \
+	COUNT=0; \
+	FAILED=0; \
+	START_TIME=$$(date +%s); \
+	while IFS=: read -r MODEL VOICE; do \
+		COUNT=$$((COUNT + 1)); \
+		printf "[%3d/%3d] Testing %-20s %-30s ... " "$$COUNT" "$$TOTAL" "$$MODEL" "$$VOICE"; \
+		if curl -s -X POST http://$(REMOTE_HOST):8000/v1/audio/speech \
+			-H "Content-Type: application/json" \
+			-d "{\"voice\":\"$$VOICE\",\"input\":\"Hydration test\"}" \
+			-o /tmp/hydrate_test/$${MODEL}_$${VOICE}.mp3 2>&1 | grep -q "error"; then \
+			echo "❌ FAILED"; \
+			FAILED=$$((FAILED + 1)); \
+		else \
+			SIZE=$$(stat -c%s /tmp/hydrate_test/$${MODEL}_$${VOICE}.mp3 2>/dev/null || echo 0); \
+			if [ "$$SIZE" -gt 1000 ]; then \
+				echo "✅ OK ($${SIZE} bytes)"; \
+			else \
+				echo "⚠️  SMALL ($${SIZE} bytes)"; \
+				FAILED=$$((FAILED + 1)); \
+			fi; \
+		fi; \
+	done < /tmp/hydrate_voices.txt; \
+	END_TIME=$$(date +%s); \
+	DURATION=$$((END_TIME - START_TIME)); \
+	echo ""; \
+	echo "🎉 Hydration complete!"; \
+	echo "   Total voices: $$TOTAL"; \
+	echo "   Successful: $$((TOTAL - FAILED))"; \
+	echo "   Failed: $$FAILED"; \
+	echo "   Duration: $${DURATION}s"; \
+	echo "   Output: /tmp/hydrate_test/"
+
+load-test:
+	@echo "🚀 Load testing TTS service with random concurrent requests..."
+	@echo "This will send 100 concurrent requests with random voices across all models"
+	@echo ""
+	@mkdir -p /tmp/load_test
+	@curl -s http://$(REMOTE_HOST):8000/v1/voices | jq -r '.data[] as $$model | $$model.voices[] | "\($$model.id):\(.)"' > /tmp/load_test_voices.txt
+	@TOTAL_VOICES=$$(wc -l < /tmp/load_test_voices.txt); \
+	REQUESTS=100; \
+	CONCURRENT=10; \
+	echo "Available voices: $$TOTAL_VOICES"; \
+	echo "Total requests: $$REQUESTS"; \
+	echo "Concurrent: $$CONCURRENT"; \
+	echo ""; \
+	START_TIME=$$(date +%s); \
+	seq 1 $$REQUESTS | xargs -P$$CONCURRENT -I{} bash -c ' \
+		LINE=$$((RANDOM % $(TOTAL_VOICES) + 1)); \
+		VOICE_SPEC=$$(sed -n "$${LINE}p" /tmp/load_test_voices.txt); \
+		MODEL=$$(echo $$VOICE_SPEC | cut -d: -f1); \
+		VOICE=$$(echo $$VOICE_SPEC | cut -d: -f2); \
+		NUM={}; \
+		START=$$(date +%s%3N); \
+		if curl -s -X POST http://$(REMOTE_HOST):8000/v1/audio/speech \
+			-H "Content-Type: application/json" \
+			-d "{\"voice\":\"$$VOICE\",\"input\":\"Load test number $$NUM\"}" \
+			-o /tmp/load_test/request_$${NUM}.mp3 2>&1; then \
+			END=$$(date +%s%3N); \
+			DURATION=$$((END - START)); \
+			SIZE=$$(stat -c%s /tmp/load_test/request_$${NUM}.mp3 2>/dev/null || echo 0); \
+			printf "[%3d] %-20s %-25s %5dms %6d bytes\n" "$$NUM" "$$MODEL" "$$VOICE" "$$DURATION" "$$SIZE"; \
+		else \
+			printf "[%3d] %-20s %-25s FAILED\n" "$$NUM" "$$MODEL" "$$VOICE"; \
+		fi \
+	'; \
+	END_TIME=$$(date +%s); \
+	DURATION=$$((END_TIME - START_TIME)); \
+	SUCCESS=$$(ls /tmp/load_test/*.mp3 2>/dev/null | wc -l); \
+	echo ""; \
+	echo "🎉 Load test complete!"; \
+	echo "   Total requests: $$REQUESTS"; \
+	echo "   Successful: $$SUCCESS"; \
+	echo "   Failed: $$((REQUESTS - SUCCESS))"; \
+	echo "   Duration: $${DURATION}s"; \
+	echo "   Avg: $$((DURATION * 1000 / REQUESTS))ms per request"; \
+	echo "   Throughput: $$((REQUESTS / DURATION)) req/s"; \
+	echo "   Output: /tmp/load_test/"
diff --git a/speech.py b/speech.py
index 85c004d..4695055 100755
--- a/speech.py
+++ b/speech.py
@@ -46,6 +46,10 @@ voice_to_model_cache = {}
 # Cached voice data for /v1/voices endpoint (loaded at startup)
 voices_cache = None
 
+# Semaphores to limit concurrent model loading (prevent thread pool exhaustion)
+silero_load_semaphore = asyncio.Semaphore(1)  # Only one Silero model load at a time
+kokoro_load_semaphore = asyncio.Semaphore(1)  # Only one Kokoro model load at a time
+
 def unload_model():
     import torch, gc
     global xtts
@@ -658,10 +662,14 @@ async def generate_speech(request: GenerateSpeechRequest):
 
         # Load Silero model if not already loaded or if language/speaker changed
         if silero_model is None or silero_speakers.get('current') != model_key:
-            logger.info(f"Loading/switching Silero model to {language}/{silero_speaker_key}")
-            # Run blocking model initialization in thread pool to avoid blocking event loop
-            silero_model = await asyncio.to_thread(silero_wrapper, language=language, speaker=silero_speaker_key, device='cpu')
-            silero_speakers['current'] = model_key
+            # Use semaphore to prevent multiple simultaneous model loads
+            async with silero_load_semaphore:
+                # Double-check after acquiring lock (another request may have loaded it)
+                if silero_model is None or silero_speakers.get('current') != model_key:
+                    logger.info(f"Loading/switching Silero model to {language}/{silero_speaker_key}")
+                    # Run blocking model initialization in thread pool to avoid blocking event loop
+                    silero_model = await asyncio.to_thread(silero_wrapper, language=language, speaker=silero_speaker_key, device='cpu')
+                    silero_speakers['current'] = model_key
 
         # Generate audio (also blocking, so run in thread pool)
         audio_data = await asyncio.to_thread(silero_model.tts, input_text, speaker_id=speaker_id)
@@ -809,4 +817,6 @@ if __name__ == "__main__":
     app.register_model('tts-1-silero')
     app.register_model('tts-1-kokoro')
 
-    uvicorn.run(app, host=args.host, port=args.port)
+    # Use multiple workers for true concurrency (each worker = separate process with own GIL)
+    # This prevents thread pool exhaustion and allows concurrent model loading
+    uvicorn.run(app, host=args.host, port=args.port, workers=4, timeout_keep_alive=300)