Make Qwen3-TTS the default engine, add CPU-only docker support

- Switch default TTS engine from Piper to Qwen3-TTS (1.7B params) - Upgrade to Python 3.12 - Add docker-compose.cpu.yml for CPU-only deployments - Improve GPU configuration with NVIDIA environment variables - Comment out optional engines (Piper, XTTS, Silero, Kokoro) in requirements - Update Makefile with local/local-cpu targets and venv support - Simplify voice_to_speaker.default.yaml for Qwen3-TTS voices - Update docs/MODELS.md with Qwen3-TTS documentation - Add git commit guidelines to CLAUDE.md
2026-01-26 10:40:29 -05:00 · 2026-01-26 10:40:29 -05:00 · b315659be6
commit b315659be6
parent 4a019cf897
9 changed files with 572 additions and 951 deletions
--- a/CLAUDE.md
+++ b/CLAUDE.md
@ -79,7 +79,13 @@ owned_by = "UncloseAI"

 **Never view or log secrets** - source them and use them.

-### 4. Documentation Requirements
+### 4. Git Commit Guidelines
+
+- **Never add AI attribution** - Do not use `Co-Authored-By: Claude` or similar in commit messages
+- Write clear, concise commit messages describing what changed and why
+- Use imperative mood ("Add feature" not "Added feature")
+
+### 5. Documentation Requirements

 When adding features, update ALL relevant docs:
 - `Makefile` help text
--- a/2
+++ b/2
@ -1,4 +1,4 @@
-FROM python:3.11-slim
+FROM python:3.12-slim

 RUN --mount=type=cache,target=/root/.cache/pip pip install -U pip

--- a/145
+++ b/145
@ -14,41 +14,48 @@ 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 hydrate load-test
+.PHONY: help deploy sync restart logs test clean stop start voices voices-qwen voices-piper voices-xtts voices-kokoro test-kokoro voices-silero test-silero voices-chatterbox test-chatterbox push-all hydrate load-test test-qwen venv venv-run local local-cpu

 help:
 	@echo "🦝 Raccoon TTS Mission - Development Commands"
 	@echo ""
-	@echo "Deployment:"
+	@echo "Quick Start (Docker with GPU):"
+	@echo "  make local       - Build and run locally with GPU support"
+	@echo "  make local-cpu   - Build and run locally (CPU only, slower)"
+	@echo "  make test        - Test Qwen3-TTS endpoint"
+	@echo "  make logs        - Tail container logs"
+	@echo ""
+	@echo "Quick Start (No Docker):"
+	@echo "  make venv        - Create Python virtual environment"
+	@echo "  make venv-run    - Run server in virtual environment"
+	@echo ""
+	@echo "Remote Deployment:"
 	@echo "  make deploy      - Full deploy: sync files, restart container"
 	@echo "  make sync        - Sync local files to remote server"
 	@echo "  make restart     - Restart the Docker container"
 	@echo ""
-	@echo "Development:"
-	@echo "  make logs        - Tail container logs"
-	@echo "  make test        - Test TTS endpoint (Piper)"
-	@echo "  make test-xtts   - Test XTTS HD endpoint"
-	@echo "  make voices      - Download all voices (Piper + XTTS)"
-	@echo "  make voices-piper - Download Piper voices only"
-	@echo "  make voices-xtts  - Download XTTS voices and samples"
-	@echo "  make voices-kokoro - Download Kokoro models"
-	@echo "  make test-kokoro   - Test Kokoro fast TTS"
-	@echo "  make voices-silero - Download Silero models (en, ru, de, es, fr)"
-	@echo "  make test-silero   - Test Silero TTS endpoint"
-	@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 "  make test        - Test TTS endpoint (Qwen3-TTS)"
+	@echo "  make test-qwen   - Test Qwen3-TTS voice cloning"
+	@echo "  make hydrate     - Test all configured voices"
+	@echo "  make load-test   - Concurrent load test"
 	@echo ""
-	@echo "Container:"
+	@echo "Other Engines (disabled by default):"
+	@echo "  make test-xtts     - Test XTTS HD endpoint"
+	@echo "  make test-kokoro   - Test Kokoro fast TTS"
+	@echo "  make test-silero   - Test Silero TTS endpoint"
+	@echo "  make voices-piper  - Download Piper voices"
+	@echo "  make voices-xtts   - Download XTTS voices"
+	@echo "  make voices-kokoro - Download Kokoro models"
+	@echo "  make voices-silero - Download Silero models"
+	@echo ""
+	@echo "Container Management:"
 	@echo "  make start       - Start Docker container"
 	@echo "  make stop        - Stop Docker container"
 	@echo "  make clean       - Stop and remove container"
 	@echo ""
 	@echo "Git:"
-	@echo "  make push-all    - Push to all git remotes (origin + github)"
+	@echo "  make push-all    - Push to all git remotes"

 sync:
 	@echo "📦 Syncing files to $(REMOTE_HOST)..."
@ -61,6 +68,78 @@ sync:
 deploy: sync restart
 	@echo "✅ Deployment complete!"

+# ============================================================================
+# Local Development (Docker)
+# ============================================================================
+
+local:
+	@echo "🐳 Building and running with GPU support..."
+	@[ -f speech.env ] || cp sample.env speech.env
+	docker compose up -d --build
+	@echo ""
+	@echo "✅ Container started! Qwen3-TTS model will download on first request (~3.4GB)"
+	@echo "   Test with: make test"
+	@echo "   View logs: make logs"
+
+local-cpu:
+	@echo "🐳 Building and running (CPU only - slower inference)..."
+	@[ -f speech.env ] || cp sample.env speech.env
+	docker compose -f docker-compose.cpu.yml up -d --build
+	@echo ""
+	@echo "✅ Container started (CPU mode)!"
+	@echo "   Note: Qwen3-TTS is ~10x slower on CPU"
+	@echo "   Test with: make test"
+
+# ============================================================================
+# Local Development (Python venv - no Docker)
+# ============================================================================
+
+VENV_DIR := .venv
+PYTHON := python3
+
+venv:
+	@echo "🐍 Creating Python virtual environment..."
+	@if [ ! -d "$(VENV_DIR)" ]; then \
+		$(PYTHON) -m venv $(VENV_DIR); \
+		echo "✅ Virtual environment created at $(VENV_DIR)"; \
+	else \
+		echo "ℹ️  Virtual environment already exists at $(VENV_DIR)"; \
+	fi
+	@echo ""
+	@echo "📦 Installing dependencies..."
+	$(VENV_DIR)/bin/pip install --upgrade pip
+	$(VENV_DIR)/bin/pip install -r requirements.txt
+	@echo ""
+	@echo "✅ Setup complete!"
+	@echo ""
+	@echo "To activate manually:"
+	@echo "  source $(VENV_DIR)/bin/activate"
+	@echo ""
+	@echo "To run the server:"
+	@echo "  make venv-run"
+	@echo ""
+	@echo "Or run directly:"
+	@echo "  $(VENV_DIR)/bin/python speech.py"
+
+venv-run:
+	@echo "🚀 Starting uncloseai-speech server..."
+	@if [ ! -d "$(VENV_DIR)" ]; then \
+		echo "❌ Virtual environment not found. Run 'make venv' first."; \
+		exit 1; \
+	fi
+	@[ -d "config" ] || mkdir -p config
+	@[ -d "voices" ] || mkdir -p voices
+	@echo ""
+	@echo "Server starting on http://localhost:8000"
+	@echo "Qwen3-TTS model will download on first request (~3.4GB)"
+	@echo ""
+	$(VENV_DIR)/bin/python speech.py
+
+venv-clean:
+	@echo "🧹 Removing virtual environment..."
+	rm -rf $(VENV_DIR)
+	@echo "✅ Virtual environment removed"
+
 restart:
 	@echo "🔄 Rebuilding and restarting container on $(REMOTE_HOST)..."
 	ssh $(REMOTE_USER)@$(REMOTE_HOST) "cd $(REMOTE_PATH) && docker compose up -d --build"
@ -81,17 +160,29 @@ logs:
 	@echo "📋 Tailing logs from $(REMOTE_HOST)..."
 	ssh $(REMOTE_USER)@$(REMOTE_HOST) "docker logs -f $(CONTAINER_NAME)"

-test:
-	@echo "🧪 Testing TTS endpoint..."
+test: test-qwen
+
+test-qwen:
+	@echo "🧪 Testing Qwen3-TTS endpoint (default model)..."
 	curl -X POST http://$(REMOTE_HOST):8000/v1/audio/speech \
 		-H "Content-Type: application/json" \
-		-d '{"model":"tts-1","voice":"alloy","input":"Raccoon mission TTS test"}' \
-		-o /tmp/raccoon_test.mp3
+		-d '{"model":"tts-1-qwen","voice":"alloy","input":"Raccoon mission TTS test with Qwen three"}' \
+		-o /tmp/qwen_test.mp3
 	@echo "✅ Test complete! Playing audio..."
-	@firefox /tmp/raccoon_test.mp3 || mpv /tmp/raccoon_test.mp3 || echo "Install firefox or mpv to play audio"
+	@firefox /tmp/qwen_test.mp3 || mpv /tmp/qwen_test.mp3 || echo "Install firefox or mpv to play audio"

-voices: voices-piper voices-xtts voices-silero
-	@echo "✅ All voices downloaded (Piper, XTTS, Silero)!"
+voices: voices-qwen
+	@echo "✅ Qwen3-TTS ready (model downloads automatically on first use)"
+
+voices-qwen:
+	@echo "🎤 Qwen3-TTS models download automatically on first use"
+	@echo "   Model: Qwen/Qwen3-TTS-12Hz-1.7B-Base (~3.4GB)"
+	@echo "   The model will be cached in /app/voices/hub/"
+	@echo ""
+	@echo "To pre-download, run: make test-qwen"
+
+voices-all: voices-qwen voices-piper voices-xtts voices-silero
+	@echo "✅ All voices downloaded (Qwen, Piper, XTTS, Silero)!"

 voices-piper:
 	@echo "🎤 Downloading all Piper voices..."
--- a/docker-compose.cpu.yml
+++ b/docker-compose.cpu.yml
@ -0,0 +1,18 @@
+# CPU-only configuration (no GPU)
+# Use this if you don't have an NVIDIA GPU
+# Note: Qwen3-TTS is ~10x slower on CPU
+
+services:
+  server:
+    build:
+      dockerfile: Dockerfile
+    image: uncloseai-speech:local
+    env_file: speech.env
+    ports:
+      - "8000:8000"
+    volumes:
+      - ./voices:/app/voices
+      - ./config:/app/config
+    environment:
+      - CUDA_VISIBLE_DEVICES=  # Disable CUDA
+    restart: unless-stopped
--- a/docker-compose.yml
+++ b/docker-compose.yml
@ -1,3 +1,7 @@
+# GPU configuration (NVIDIA CUDA)
+# Requires: nvidia-container-toolkit installed on host
+# Install: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html
+
 services:
  server:
    build:
@ -9,13 +13,16 @@ services:
    volumes:
      - ./voices:/app/voices
      - ./config:/app/config
-    # To install as a service
+    environment:
+      - NVIDIA_VISIBLE_DEVICES=all
+      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    restart: unless-stopped
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
-              #device_ids: ['0', '1'] # Select a gpu, or
+              # Uncomment to select specific GPU(s):
+              # device_ids: ['0']
              count: all
              capabilities: [gpu]
--- a/docs/MODELS.md
+++ b/docs/MODELS.md
@ -2,6 +2,22 @@

 **Raccoon Mission:** Rescue abandoned open-source TTS models and integrate them into uncloseai-speech

+## Default Model: Qwen3-TTS
+
+**🎯 Qwen3-TTS is now the default and only enabled model.**
+
+All other models (Piper, XTTS, Silero, Kokoro) are disabled by default. To enable them, uncomment their sections in `voice_to_speaker.yaml`.
+
+### Quick Start
+```bash
+# Test Qwen3-TTS (default)
+make test
+
+# The model downloads automatically on first use (~3.4GB)
+```
+
+---
+
 ## Documentation Index

 ### Comprehensive Research
@ -33,7 +49,83 @@ Each model has detailed documentation covering technical specs, integration stat

 ## Currently Integrated

-### 1. Piper TTS ✅
+### 0. Qwen3-TTS ✅ (DEFAULT)
+
+**Status:** INTEGRATED as tts-1-qwen (DEFAULT MODEL)
+**Project:** Qwen/Qwen3-TTS (Alibaba, actively maintained)
+**License:** Apache 2.0
+**Model:** Qwen3-TTS-12Hz-1.7B-Base
+
+**Why Default:**
+- State-of-the-art quality with 1.7B parameters
+- Extremely low latency (97ms first packet)
+- Voice cloning from 3-second samples
+- 10 languages: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian
+- Apache 2.0 license (commercial-friendly)
+- Actively maintained by Alibaba
+
+**Features:**
+- Universal end-to-end architecture (no cascading errors)
+- 12Hz acoustic tokenizer for efficient compression
+- Dual-track streaming/non-streaming generation
+- High-fidelity speech reconstruction
+- Natural language instruction control
+- Supports both GPU and CPU inference
+
+**Model Specs:**
+- Parameters: 1.7B
+- Sample Rate: ~24kHz
+- Input: Text + Reference Audio (3+ seconds)
+- Languages: 10 (zh, en, ja, ko, de, fr, ru, pt, es, it)
+- Size: ~3.4GB
+
+**Model Source:**
+- HuggingFace: `Qwen/Qwen3-TTS-12Hz-1.7B-Base`
+- Auto-downloaded on first use via huggingface-hub
+- Cached in `/app/voices/hub/`
+
+**Integration:**
+- Used for `tts-1-qwen` model (default)
+- Voice cloning with reference audio + transcript
+- Pre-configured with Qwen's demo voice
+
+**Example Config:**
+```yaml
+tts-1-qwen:
+  alloy:
+    ref_audio: https://example.com/reference.wav
+    ref_text: "The exact text spoken in the reference audio"
+    language: English
+```
+
+**Custom Voice Setup:**
+1. Record 3+ seconds of clear speech
+2. Transcribe the audio exactly
+3. Add to `voice_to_speaker.yaml`:
+```yaml
+tts-1-qwen:
+  my_voice:
+    ref_audio: voices/my_voice_sample.wav
+    ref_text: "Hello, this is my voice sample for cloning."
+    language: English
+```
+
+**Makefile Targets:**
+```bash
+make test        # Test Qwen3-TTS (default)
+make test-qwen   # Test Qwen3-TTS explicitly
+```
+
+**Hardware Requirements:**
+- GPU: NVIDIA with 8GB+ VRAM (recommended)
+- CPU: Works but slower (~10x)
+- FlashAttention 2 recommended for lower memory
+
+**Raccoon Priority:** ⭐⭐⭐⭐⭐ (State-of-the-art, actively maintained, Apache 2.0)
+
+---
+
+### 1. Piper TTS (disabled by default) ✅

 > 📖 **See [detailed documentation](models/piper-tts.md)** for comprehensive technical specs and integration guide

@ -505,8 +597,8 @@ The following models have detailed documentation but are not yet integrated or p

 ---

-**Last Updated:** 2025-11-09
-**Raccoon Status:** 🦝 4 models rescued! Silero and Kokoro TTS integrated successfully
-**Integration Status:** ✅ Piper (55 voices), XTTS (8 voices), Silero (148 voices), Kokoro (34 voices) | 🎯 Next: Chatterbox, StyleTTS2
-**API Endpoints:** tts-1, tts-1-hd, tts-1-silero, tts-1-kokoro | /v1/models for discovery
-**Documentation Status:** 📚 10 models fully documented, 1 comprehensive research overview
+**Last Updated:** 2026-01-26
+**Raccoon Status:** 🦝 5 models rescued! Qwen3-TTS is now the default model
+**Integration Status:** ✅ Qwen3-TTS (default, unlimited voices via cloning) | Disabled: Piper (55), XTTS (8), Silero (148), Kokoro (34)
+**API Endpoints:** tts-1-qwen (default) | Others available: tts-1, tts-1-hd, tts-1-silero, tts-1-kokoro
+**Documentation Status:** 📚 11 models fully documented, 1 comprehensive research overview
--- a/requirements.txt
+++ b/requirements.txt
@ -1,28 +1,31 @@
 fastapi
 uvicorn
 loguru
+# Qwen3-TTS - state-of-the-art TTS with voice cloning (Apache 2.0)
+# 1.7B params, 10 languages, 97ms latency, 12Hz tokenizer
+qwen-tts>=0.1.0
 # OHF-Voice fork doesn't have installable Python package yet
 # Stick with PyPI piper-tts but use absolute paths in config
-piper-tts>=1.2.0
+# piper-tts>=1.2.0
 # 🦝 RACCOON TODO: Create our own PyPI package from OHF-Voice fork
 # git+https://github.com/OHF-Voice/piper1-gpl.git@v1.3.0#subdirectory=src/python_run
-coqui-tts[languages]
+# coqui-tts[languages]
 # Silero TTS - actively maintained, small efficient models
 # Note: Silero models are loaded via torch.hub, no package install needed
 # Models: ~50-100MB each, CPU-friendly, real-time capable
-omegaconf  # Required by Silero TTS
+# omegaconf  # Required by Silero TTS
 # Chatterbox - emotion control, 23 languages (Resemble AI)
 # Install from git since no PyPI package exists yet
 # 🦝 RACCOON NOTE: Disabled due to dependency conflict with Coqui TTS
 # gradio 5.44.1 requires typer<1.0 and >=0.12, but spacy 3.6.x requires typer<0.10.0
 # TODO: Test Chatterbox in isolated environment or wait for dependency updates
 # git+https://github.com/resemble-ai/chatterbox.git
-langdetect
+# langdetect
 pyyaml
 # Kokoro TTS - fast decoder-only architecture
 # Lightweight decoder-only TTS, 82M params, 24kHz output
-kokoro>=0.9.2
-soundfile  # Required by Kokoro for audio output
+# kokoro>=0.9.2
+soundfile  # Required by Qwen3-TTS and Kokoro for audio output
 transformers>=4.35.0
 # Hugging Face Hub for model downloads
 huggingface-hub[cli]
--- a/speech.py
+++ b/speech.py
@ -38,6 +38,20 @@ except ImportError:
    split_sentence = None
    detect = None

+# Try to import Qwen3-TTS dependencies
+try:
+    import torch
+    from qwen_tts import Qwen3TTSModel
+    QWEN_TTS_AVAILABLE = True
+except ImportError:
+    QWEN_TTS_AVAILABLE = False
+    if torch is None:
+        try:
+            import torch
+        except ImportError:
+            torch = None
+    Qwen3TTSModel = None
+
@contextlib.asynccontextmanager
 async def lifespan(app):
    # Startup: Initialize voice caches in each worker process
@ -89,6 +103,10 @@ async def lifespan(app):
                model_info["engine"] = "kokoro"
                model_info["description"] = "Lightweight decoder-only TTS (82M params)"
                model_info["sample_rate"] = 24000
+            elif model_id == 'tts-1-qwen':
+                model_info["engine"] = "qwen3-tts"
+                model_info["description"] = "State-of-the-art TTS with voice cloning (1.7B params, 10 languages)"
+                model_info["sample_rate"] = 24000

            models_data.append(model_info)

@ -125,6 +143,8 @@ silero_model = None
 silero_speakers = {}
 kokoro_pipeline = None
 kokoro_lang = None
+qwen_model = None
+qwen_voice_prompts = {}  # Cache for voice clone prompts

 # Default args for worker processes (will be overridden in __main__)
 class DefaultArgs:
@ -159,6 +179,7 @@ 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
+qwen_load_semaphore = asyncio.Semaphore(1)    # Only one Qwen model load at a time

 def unload_model():
    import torch, gc
@ -350,6 +371,126 @@ class kokoro_wrapper():
            logger.error(f"Kokoro TTS generation failed: {e}")
            raise

+class qwen3_wrapper():
+    """Wrapper for Qwen3-TTS model
+
+    Qwen3-TTS is a state-of-the-art TTS with voice cloning:
+    - 1.7B parameters, 12Hz tokenizer
+    - 10 languages: zh, en, ja, ko, de, fr, ru, pt, es, it
+    - 97ms first-packet latency
+    - 3-second rapid voice cloning
+    Output: Variable sample rate (typically 24kHz)
+    """
+    def __init__(self, model_name='Qwen/Qwen3-TTS-12Hz-1.7B-Base', device='cuda', dtype=None):
+        self.model_name = model_name
+        self.device = device
+        self.sample_rate = None  # Set after first generation
+        self.voice_prompts = {}  # Cache for reusable voice clone prompts
+
+        logger.info(f"Loading Qwen3-TTS model '{model_name}' on device '{device}'")
+
+        try:
+            import torch
+            from qwen_tts import Qwen3TTSModel
+
+            # Determine dtype
+            if dtype is None:
+                if device == 'cuda' and torch.cuda.is_available():
+                    dtype = torch.bfloat16
+                else:
+                    dtype = torch.float32
+
+            # Try to use flash attention if available
+            try:
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_name,
+                    device_map=device,
+                    dtype=dtype,
+                    attn_implementation="flash_attention_2",
+                )
+                logger.info(f"Loaded Qwen3-TTS with FlashAttention 2")
+            except Exception as fa_error:
+                logger.warning(f"FlashAttention 2 not available ({fa_error}), using default attention")
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_name,
+                    device_map=device,
+                    dtype=dtype,
+                )
+
+            logger.info(f"Successfully loaded Qwen3-TTS model on {device}")
+        except Exception as e:
+            logger.error(f"Failed to load Qwen3-TTS model: {e}")
+            raise
+
+    def create_voice_prompt(self, ref_audio, ref_text, x_vector_only_mode=False):
+        """Create a reusable voice clone prompt from reference audio/text.
+
+        Args:
+            ref_audio: Path to reference audio file, URL, or (numpy_array, sample_rate) tuple
+            ref_text: Text spoken in the reference audio
+            x_vector_only_mode: If True, use only speaker embedding (faster but lower quality)
+
+        Returns:
+            Voice clone prompt items for reuse
+        """
+        logger.info(f"Creating voice clone prompt from ref_audio={ref_audio}, ref_text={ref_text[:50]}...")
+        return self.model.create_voice_clone_prompt(
+            ref_audio=ref_audio,
+            ref_text=ref_text,
+            x_vector_only_mode=x_vector_only_mode,
+        )
+
+    def tts(self, text, language='English', ref_audio=None, ref_text=None, voice_prompt=None):
+        """Generate speech from text using voice cloning.
+
+        Args:
+            text: Text to synthesize (string or list of strings)
+            language: Language of the text
+            ref_audio: Path/URL to reference audio (if voice_prompt not provided)
+            ref_text: Text in reference audio (if voice_prompt not provided)
+            voice_prompt: Pre-computed voice clone prompt (for efficiency)
+
+        Returns:
+            Audio data as bytes (float32 PCM)
+        """
+        import numpy as np
+
+        logger.info(f"Qwen3-TTS generating: text length={len(text)}, language={language}")
+
+        try:
+            if voice_prompt is not None:
+                # Use pre-computed voice prompt
+                wavs, sr = self.model.generate_voice_clone(
+                    text=text,
+                    language=language,
+                    voice_clone_prompt=voice_prompt,
+                )
+            elif ref_audio is not None and ref_text is not None:
+                # Generate with inline reference
+                wavs, sr = self.model.generate_voice_clone(
+                    text=text,
+                    language=language,
+                    ref_audio=ref_audio,
+                    ref_text=ref_text,
+                )
+            else:
+                raise ValueError("Either voice_prompt or (ref_audio + ref_text) must be provided")
+
+            self.sample_rate = sr
+            logger.info(f"Qwen3-TTS generated {len(wavs)} audio segment(s) at {sr}Hz")
+
+            # Concatenate all wav segments and convert to bytes
+            if len(wavs) > 1:
+                full_audio = np.concatenate(wavs)
+            else:
+                full_audio = wavs[0]
+
+            return full_audio.astype(np.float32).tobytes()
+
+        except Exception as e:
+            logger.error(f"Qwen3-TTS generation failed: {e}")
+            raise
+
 def default_exists(filename: str):
    if not os.path.exists(filename):
        fpath, ext = os.path.splitext(filename)
@ -462,33 +603,41 @@ def build_ffmpeg_args(response_format, input_format, sample_rate):
 async def list_models():
    """List all available TTS models (OpenAI-compatible format)"""
    # Return minimal OpenAI-compatible model list (no extra fields)
+    # Only tts-1-qwen is enabled by default
    return {
        "object": "list",
        "data": [
            {
-                "id": "tts-1",
-                "object": "model",
-                "created": 1700000000,
-                "owned_by": "uncloseai"
-            },
-            {
-                "id": "tts-1-hd",
-                "object": "model",
-                "created": 1700000000,
-                "owned_by": "uncloseai"
-            },
-            {
-                "id": "tts-1-silero",
-                "object": "model",
-                "created": 1700000000,
-                "owned_by": "uncloseai"
-            },
-            {
-                "id": "tts-1-kokoro",
+                "id": "tts-1-qwen",
                "object": "model",
                "created": 1700000000,
                "owned_by": "uncloseai"
            }
+            # Other models disabled by default:
+            # {
+            #     "id": "tts-1",
+            #     "object": "model",
+            #     "created": 1700000000,
+            #     "owned_by": "uncloseai"
+            # },
+            # {
+            #     "id": "tts-1-hd",
+            #     "object": "model",
+            #     "created": 1700000000,
+            #     "owned_by": "uncloseai"
+            # },
+            # {
+            #     "id": "tts-1-silero",
+            #     "object": "model",
+            #     "created": 1700000000,
+            #     "owned_by": "uncloseai"
+            # },
+            # {
+            #     "id": "tts-1-kokoro",
+            #     "object": "model",
+            #     "created": 1700000000,
+            #     "owned_by": "uncloseai"
+            # }
        ]
    }

@ -542,6 +691,10 @@ async def list_voices():
                model_info["engine"] = "kokoro"
                model_info["description"] = "Lightweight decoder-only TTS (82M params)"
                model_info["sample_rate"] = 24000
+            elif model_id == 'tts-1-qwen':
+                model_info["engine"] = "qwen3-tts"
+                model_info["description"] = "State-of-the-art TTS with voice cloning (1.7B params, 10 languages)"
+                model_info["sample_rate"] = 24000

            models_data.append(model_info)

@ -597,6 +750,8 @@ async def generate_speech(request: GenerateSpeechRequest):
            media_type = "audio/pcm;rate=48000"
        elif model == 'tts-1-kokoro': # kokoro
            media_type = "audio/pcm;rate=24000"
+        elif model == 'tts-1-qwen': # qwen3-tts
+            media_type = "audio/pcm;rate=24000"
    else:
        raise BadRequestError(f"Invalid response_format: '{response_format}'", param='response_format')

@ -912,9 +1067,70 @@ async def generate_speech(request: GenerateSpeechRequest):
        out_writer_worker = threading.Thread(target=out_writer, daemon=True)
        out_writer_worker.start()

+        return StreamingResponse(content=ffmpeg_proc.stdout, media_type=media_type)
+    # Use Qwen3-TTS for tts-1-qwen
+    elif model == 'tts-1-qwen':
+        global qwen_model, qwen_voice_prompts
+
+        if not QWEN_TTS_AVAILABLE:
+            raise ServiceUnavailableError("Qwen3-TTS is not available. Install with: pip install qwen-tts")
+
+        voice_map = map_voice_to_speaker(voice, 'tts-1-qwen')
+        ref_audio = voice_map.get('ref_audio')
+        ref_text = voice_map.get('ref_text')
+        language = voice_map.get('language', 'English')
+
+        # Load Qwen model if not already loaded
+        if qwen_model is None:
+            async with qwen_load_semaphore:
+                if qwen_model is None:
+                    device = args.xtts_device if args.xtts_device != 'none' else 'cpu'
+                    logger.info(f"Loading Qwen3-TTS model on device '{device}'")
+                    qwen_model = await asyncio.to_thread(
+                        qwen3_wrapper,
+                        model_name='Qwen/Qwen3-TTS-12Hz-1.7B-Base',
+                        device=device
+                    )
+
+        # Create or retrieve cached voice prompt
+        voice_prompt = None
+        if ref_audio and ref_text:
+            cache_key = f"{voice}_{ref_audio}"
+            if cache_key not in qwen_voice_prompts:
+                logger.info(f"Creating voice prompt for '{voice}'")
+                qwen_voice_prompts[cache_key] = await asyncio.to_thread(
+                    qwen_model.create_voice_prompt,
+                    ref_audio=ref_audio,
+                    ref_text=ref_text
+                )
+            voice_prompt = qwen_voice_prompts[cache_key]
+        else:
+            raise BadRequestError(f"Voice '{voice}' requires ref_audio and ref_text configuration", param='voice')
+
+        # Generate audio
+        audio_data = await asyncio.to_thread(
+            qwen_model.tts,
+            text=input_text,
+            language=language,
+            voice_prompt=voice_prompt
+        )
+
+        # Qwen outputs float32 PCM at ~24kHz (sample rate from model)
+        sample_rate = str(qwen_model.sample_rate or 24000)
+        ffmpeg_args = build_ffmpeg_args(response_format, input_format="f32le", sample_rate=sample_rate)
+
+        # Apply speed adjustment if needed
+        if speed != 1.0:
+            ffmpeg_args.extend(["-af", f"atempo={speed}"])
+
+        ffmpeg_args.extend(["-"])
+        ffmpeg_proc = subprocess.Popen(ffmpeg_args, stdin=subprocess.PIPE, stdout=subprocess.PIPE)
+        ffmpeg_proc.stdin.write(audio_data)
+        ffmpeg_proc.stdin.close()
+
        return StreamingResponse(content=ffmpeg_proc.stdout, media_type=media_type)
    else:
-        raise BadRequestError("No such model, must be tts-1, tts-1-hd, tts-1-silero, or tts-1-kokoro.", param='model')
+        raise BadRequestError("No such model, must be tts-1-qwen (default), tts-1, tts-1-hd, tts-1-silero, or tts-1-kokoro.", param='model')

 if __name__ == "__main__":
    parser = argparse.ArgumentParser(
@ -941,10 +1157,13 @@ if __name__ == "__main__":
    elif args.preload:
        xtts = xtts_wrapper(args.preload, device=args.xtts_device, unload_timer=args.unload_timer)

-    app.register_model('tts-1')
-    app.register_model('tts-1-hd')
-    app.register_model('tts-1-silero')
-    app.register_model('tts-1-kokoro')
+    # Register only Qwen by default (other models disabled)
+    app.register_model('tts-1-qwen')
+    # To enable other models, uncomment below:
+    # app.register_model('tts-1')
+    # app.register_model('tts-1-hd')
+    # app.register_model('tts-1-silero')
+    # app.register_model('tts-1-kokoro')

    # Use multiple workers for true concurrency (each worker = separate process with own GIL)
    # This prevents thread pool exhaustion and allows concurrent model loading
--- a/voice_to_speaker.default.yaml
+++ b/voice_to_speaker.default.yaml
@ -1,892 +1,77 @@
-tts-1:
-  # OpenAI-compatible voice aliases (backward compatibility)
-  alloy:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 79 # 64, 79, 80, 101, 130
-  echo:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 134 # 52, 102, 134
-  fable:
-    model: /app/voices/en/en_GB/northern_english_male/medium/en_GB-northern_english_male-medium.onnx
-    speaker: # default speaker
-  onyx:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 159 # 55, 90, 132, 136, 137, 159
-  nova:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 107 # 57, 61, 107, 150, 162
-  shimmer:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 163
+# uncloseai-speech Voice Configuration
+# Only tts-1-qwen is enabled by default

-  # English US voices - all available Piper models
-  # libritts_r has 904 speakers (multi-speaker model)
-  en_us_libritts_r_0:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 0
-  en_us_libritts_r_52:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 52
-  en_us_libritts_r_55:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 55
-  en_us_libritts_r_57:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 57
-  en_us_libritts_r_61:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 61
-  en_us_libritts_r_64:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 64
-  en_us_libritts_r_79:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 79
-  en_us_libritts_r_80:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 80
-  en_us_libritts_r_90:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 90
-  en_us_libritts_r_101:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 101
-  en_us_libritts_r_102:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 102
-  en_us_libritts_r_107:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 107
-  en_us_libritts_r_130:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 130
-  en_us_libritts_r_132:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 132
-  en_us_libritts_r_134:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 134
-  en_us_libritts_r_136:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 136
-  en_us_libritts_r_137:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 137
-  en_us_libritts_r_150:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 150
-  en_us_libritts_r_159:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 159
-  en_us_libritts_r_162:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 162
-  en_us_libritts_r_163:
-    model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
-    speaker: 163
+tts-1-qwen:
+  # OpenAI-compatible voice aliases
+  # Each voice requires ref_audio (reference audio) and ref_text (transcript of the audio)
+  # Language: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian

-  # Single-speaker Piper voices (to be downloaded)
-  en_us_amy:
-    model: /app/voices/en/en_US/amy/medium/en_US-amy-medium.onnx
-    speaker: # default
-  en_us_arctic:
-    model: /app/voices/en/en_US/arctic/medium/en_US-arctic-medium.onnx
-    speaker: # default
-  en_us_bryce:
-    model: /app/voices/en/en_US/bryce/medium/en_US-bryce-medium.onnx
-    speaker: # default
-  en_us_danny:
-    model: /app/voices/en/en_US/danny/low/en_US-danny-low.onnx
-    speaker: # default
-  en_us_hfc_female:
-    model: /app/voices/en/en_US/hfc_female/medium/en_US-hfc_female-medium.onnx
-    speaker: # default
-  en_us_hfc_male:
-    model: /app/voices/en/en_US/hfc_male/medium/en_US-hfc_male-medium.onnx
-    speaker: # default
-  en_us_joe:
-    model: /app/voices/en/en_US/joe/medium/en_US-joe-medium.onnx
-    speaker: # default
-  en_us_john:
-    model: /app/voices/en/en_US/john/medium/en_US-john-medium.onnx
-    speaker: # default
-  en_us_kathleen:
-    model: /app/voices/en/en_US/kathleen/low/en_US-kathleen-low.onnx
-    speaker: # default
-  en_us_kristin:
-    model: /app/voices/en/en_US/kristin/medium/en_US-kristin-medium.onnx
-    speaker: # default
-  en_us_kusal:
-    model: /app/voices/en/en_US/kusal/medium/en_US-kusal-medium.onnx
-    speaker: # default
-  en_us_l2arctic:
-    model: /app/voices/en/en_US/l2arctic/medium/en_US-l2arctic-medium.onnx
-    speaker: # default (multi-speaker model)
-  en_us_lessac:
-    model: /app/voices/en/en_US/lessac/medium/en_US-lessac-medium.onnx
-    speaker: # default (multi-speaker model)
-  en_us_libritts:
-    model: /app/voices/en/en_US/libritts/high/en_US-libritts-high.onnx
-    speaker: # default (multi-speaker model)
-  en_us_ljspeech:
-    model: /app/voices/en/en_US/ljspeech/medium/en_US-ljspeech-medium.onnx
-    speaker: # default
-  en_us_norman:
-    model: /app/voices/en/en_US/norman/medium/en_US-norman-medium.onnx
-    speaker: # default
-  en_us_reza_ibrahim:
-    model: /app/voices/en/en_US/reza_ibrahim/medium/en_US-reza_ibrahim-medium.onnx
-    speaker: # default
-  en_us_ryan:
-    model: /app/voices/en/en_US/ryan/high/en_US-ryan-high.onnx
-    speaker: # default
-  en_us_sam:
-    model: /app/voices/en/en_US/sam/medium/en_US-sam-medium.onnx
-    speaker: # default
+  # Default voice - using Qwen's example clone audio
+  alloy:
+    ref_audio: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav
+    ref_text: "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it!"
+    language: English

-  # English GB voices
-  en_gb_alan:
-    model: /app/voices/en/en_GB/alan/medium/en_GB-alan-medium.onnx
-    speaker: # default
-  en_gb_alba:
-    model: /app/voices/en/en_GB/alba/medium/en_GB-alba-medium.onnx
-    speaker: # default
-  en_gb_aru:
-    model: /app/voices/en/en_GB/aru/medium/en_GB-aru-medium.onnx
-    speaker: # default (multi-speaker model)
-  en_gb_cori:
-    model: /app/voices/en/en_GB/cori/medium/en_GB-cori-medium.onnx
-    speaker: # default (multi-speaker model)
-  en_gb_jenny_dioco:
-    model: /app/voices/en/en_GB/jenny_dioco/medium/en_GB-jenny_dioco-medium.onnx
-    speaker: # default
-  en_gb_northern_english_male:
-    model: /app/voices/en/en_GB/northern_english_male/medium/en_GB-northern_english_male-medium.onnx
-    speaker: # default
-  en_gb_semaine:
-    model: /app/voices/en/en_GB/semaine/medium/en_GB-semaine-medium.onnx
-    speaker: # default
-  en_gb_southern_english_female:
-    model: /app/voices/en/en_GB/southern_english_female/low/en_GB-southern_english_female-low.onnx
-    speaker: # default
-  en_gb_vctk:
-    model: /app/voices/en/en_GB/vctk/medium/en_GB-vctk-medium.onnx
-    speaker: # default (multi-speaker model)
-tts-1-hd:
-  alloy-alt:
-    model: xtts
-    speaker: voices/alloy-alt.wav
-  alloy:
-    model: xtts
-    speaker: voices/alloy.wav
+  # Echo - same sample, different name for compatibility
  echo:
-    model: xtts
-    speaker: voices/echo.wav
+    ref_audio: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav
+    ref_text: "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it!"
+    language: English
+
+  # Fable - same sample
  fable:
-    model: xtts
-    speaker: voices/fable.wav
+    ref_audio: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav
+    ref_text: "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it!"
+    language: English
+
+  # Onyx - same sample
  onyx:
-    model: xtts
-    speaker: voices/onyx.wav
+    ref_audio: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav
+    ref_text: "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it!"
+    language: English
+
+  # Nova - same sample
  nova:
-    model: xtts
-    speaker: voices/nova.wav
+    ref_audio: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav
+    ref_text: "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it!"
+    language: English
+
+  # Shimmer - same sample
  shimmer:
-    model: xtts
-    speaker: voices/shimmer.wav
-  me:
-    model: xtts_v2.0.2 # you can specify an older xtts version
-    speaker: voices/me.wav # this could be you
-    language: auto
-    enable_text_splitting: True
-    length_penalty: 1.0
-    repetition_penalty: 10
-    speed: 1.0
-    temperature: 0.75
-    top_k: 50
-    top_p: 0.85
-    comment: You can add a comment here also, which will be persistent and otherwise ignored.
-tts-1-silero:
-  # All 118 Silero v3_en voices with proper names
-  # NOTE: Use native Silero voice names (en_0, en_1, etc.) for Silero
-  # OpenAI voice names (alloy, echo, etc.) are reserved for tts-1 and tts-1-hd only
-  en_0:
-    language: en
-    speaker: en_0
-    silero_speaker: v3_en
-  en_1:
-    language: en
-    speaker: en_1
-    silero_speaker: v3_en
-  en_2:
-    language: en
-    speaker: en_2
-    silero_speaker: v3_en
-  en_3:
-    language: en
-    speaker: en_3
-    silero_speaker: v3_en
-  en_4:
-    language: en
-    speaker: en_4
-    silero_speaker: v3_en
-  en_5:
-    language: en
-    speaker: en_5
-    silero_speaker: v3_en
-  en_6:
-    language: en
-    speaker: en_6
-    silero_speaker: v3_en
-  en_7:
-    language: en
-    speaker: en_7
-    silero_speaker: v3_en
-  en_8:
-    language: en
-    speaker: en_8
-    silero_speaker: v3_en
-  en_9:
-    language: en
-    speaker: en_9
-    silero_speaker: v3_en
-  en_10:
-    language: en
-    speaker: en_10
-    silero_speaker: v3_en
-  en_11:
-    language: en
-    speaker: en_11
-    silero_speaker: v3_en
-  en_12:
-    language: en
-    speaker: en_12
-    silero_speaker: v3_en
-  en_13:
-    language: en
-    speaker: en_13
-    silero_speaker: v3_en
-  en_14:
-    language: en
-    speaker: en_14
-    silero_speaker: v3_en
-  en_15:
-    language: en
-    speaker: en_15
-    silero_speaker: v3_en
-  en_16:
-    language: en
-    speaker: en_16
-    silero_speaker: v3_en
-  en_17:
-    language: en
-    speaker: en_17
-    silero_speaker: v3_en
-  en_18:
-    language: en
-    speaker: en_18
-    silero_speaker: v3_en
-  en_19:
-    language: en
-    speaker: en_19
-    silero_speaker: v3_en
-  en_20:
-    language: en
-    speaker: en_20
-    silero_speaker: v3_en
-  en_21:
-    language: en
-    speaker: en_21
-    silero_speaker: v3_en
-  en_22:
-    language: en
-    speaker: en_22
-    silero_speaker: v3_en
-  en_23:
-    language: en
-    speaker: en_23
-    silero_speaker: v3_en
-  en_24:
-    language: en
-    speaker: en_24
-    silero_speaker: v3_en
-  en_25:
-    language: en
-    speaker: en_25
-    silero_speaker: v3_en
-  en_26:
-    language: en
-    speaker: en_26
-    silero_speaker: v3_en
-  en_27:
-    language: en
-    speaker: en_27
-    silero_speaker: v3_en
-  en_28:
-    language: en
-    speaker: en_28
-    silero_speaker: v3_en
-  en_29:
-    language: en
-    speaker: en_29
-    silero_speaker: v3_en
-  en_30:
-    language: en
-    speaker: en_30
-    silero_speaker: v3_en
-  en_31:
-    language: en
-    speaker: en_31
-    silero_speaker: v3_en
-  en_32:
-    language: en
-    speaker: en_32
-    silero_speaker: v3_en
-  en_33:
-    language: en
-    speaker: en_33
-    silero_speaker: v3_en
-  en_34:
-    language: en
-    speaker: en_34
-    silero_speaker: v3_en
-  en_35:
-    language: en
-    speaker: en_35
-    silero_speaker: v3_en
-  en_36:
-    language: en
-    speaker: en_36
-    silero_speaker: v3_en
-  en_37:
-    language: en
-    speaker: en_37
-    silero_speaker: v3_en
-  en_38:
-    language: en
-    speaker: en_38
-    silero_speaker: v3_en
-  en_39:
-    language: en
-    speaker: en_39
-    silero_speaker: v3_en
-  en_40:
-    language: en
-    speaker: en_40
-    silero_speaker: v3_en
-  en_41:
-    language: en
-    speaker: en_41
-    silero_speaker: v3_en
-  en_42:
-    language: en
-    speaker: en_42
-    silero_speaker: v3_en
-  en_43:
-    language: en
-    speaker: en_43
-    silero_speaker: v3_en
-  en_44:
-    language: en
-    speaker: en_44
-    silero_speaker: v3_en
-  en_45:
-    language: en
-    speaker: en_45
-    silero_speaker: v3_en
-  en_46:
-    language: en
-    speaker: en_46
-    silero_speaker: v3_en
-  en_47:
-    language: en
-    speaker: en_47
-    silero_speaker: v3_en
-  en_48:
-    language: en
-    speaker: en_48
-    silero_speaker: v3_en
-  en_49:
-    language: en
-    speaker: en_49
-    silero_speaker: v3_en
-  en_50:
-    language: en
-    speaker: en_50
-    silero_speaker: v3_en
-  en_51:
-    language: en
-    speaker: en_51
-    silero_speaker: v3_en
-  en_52:
-    language: en
-    speaker: en_52
-    silero_speaker: v3_en
-  en_53:
-    language: en
-    speaker: en_53
-    silero_speaker: v3_en
-  en_54:
-    language: en
-    speaker: en_54
-    silero_speaker: v3_en
-  en_55:
-    language: en
-    speaker: en_55
-    silero_speaker: v3_en
-  en_56:
-    language: en
-    speaker: en_56
-    silero_speaker: v3_en
-  en_57:
-    language: en
-    speaker: en_57
-    silero_speaker: v3_en
-  en_58:
-    language: en
-    speaker: en_58
-    silero_speaker: v3_en
-  en_59:
-    language: en
-    speaker: en_59
-    silero_speaker: v3_en
-  en_60:
-    language: en
-    speaker: en_60
-    silero_speaker: v3_en
-  en_61:
-    language: en
-    speaker: en_61
-    silero_speaker: v3_en
-  en_62:
-    language: en
-    speaker: en_62
-    silero_speaker: v3_en
-  en_63:
-    language: en
-    speaker: en_63
-    silero_speaker: v3_en
-  en_64:
-    language: en
-    speaker: en_64
-    silero_speaker: v3_en
-  en_65:
-    language: en
-    speaker: en_65
-    silero_speaker: v3_en
-  en_66:
-    language: en
-    speaker: en_66
-    silero_speaker: v3_en
-  en_67:
-    language: en
-    speaker: en_67
-    silero_speaker: v3_en
-  en_68:
-    language: en
-    speaker: en_68
-    silero_speaker: v3_en
-  en_69:
-    language: en
-    speaker: en_69
-    silero_speaker: v3_en
-  en_70:
-    language: en
-    speaker: en_70
-    silero_speaker: v3_en
-  en_71:
-    language: en
-    speaker: en_71
-    silero_speaker: v3_en
-  en_72:
-    language: en
-    speaker: en_72
-    silero_speaker: v3_en
-  en_73:
-    language: en
-    speaker: en_73
-    silero_speaker: v3_en
-  en_74:
-    language: en
-    speaker: en_74
-    silero_speaker: v3_en
-  en_75:
-    language: en
-    speaker: en_75
-    silero_speaker: v3_en
-  en_76:
-    language: en
-    speaker: en_76
-    silero_speaker: v3_en
-  en_77:
-    language: en
-    speaker: en_77
-    silero_speaker: v3_en
-  en_78:
-    language: en
-    speaker: en_78
-    silero_speaker: v3_en
-  en_79:
-    language: en
-    speaker: en_79
-    silero_speaker: v3_en
-  en_80:
-    language: en
-    speaker: en_80
-    silero_speaker: v3_en
-  en_81:
-    language: en
-    speaker: en_81
-    silero_speaker: v3_en
-  en_82:
-    language: en
-    speaker: en_82
-    silero_speaker: v3_en
-  en_83:
-    language: en
-    speaker: en_83
-    silero_speaker: v3_en
-  en_84:
-    language: en
-    speaker: en_84
-    silero_speaker: v3_en
-  en_85:
-    language: en
-    speaker: en_85
-    silero_speaker: v3_en
-  en_86:
-    language: en
-    speaker: en_86
-    silero_speaker: v3_en
-  en_87:
-    language: en
-    speaker: en_87
-    silero_speaker: v3_en
-  en_88:
-    language: en
-    speaker: en_88
-    silero_speaker: v3_en
-  en_89:
-    language: en
-    speaker: en_89
-    silero_speaker: v3_en
-  en_90:
-    language: en
-    speaker: en_90
-    silero_speaker: v3_en
-  en_91:
-    language: en
-    speaker: en_91
-    silero_speaker: v3_en
-  en_92:
-    language: en
-    speaker: en_92
-    silero_speaker: v3_en
-  en_93:
-    language: en
-    speaker: en_93
-    silero_speaker: v3_en
-  en_94:
-    language: en
-    speaker: en_94
-    silero_speaker: v3_en
-  en_95:
-    language: en
-    speaker: en_95
-    silero_speaker: v3_en
-  en_96:
-    language: en
-    speaker: en_96
-    silero_speaker: v3_en
-  en_97:
-    language: en
-    speaker: en_97
-    silero_speaker: v3_en
-  en_98:
-    language: en
-    speaker: en_98
-    silero_speaker: v3_en
-  en_99:
-    language: en
-    speaker: en_99
-    silero_speaker: v3_en
-  en_100:
-    language: en
-    speaker: en_100
-    silero_speaker: v3_en
-  en_101:
-    language: en
-    speaker: en_101
-    silero_speaker: v3_en
-  en_102:
-    language: en
-    speaker: en_102
-    silero_speaker: v3_en
-  en_103:
-    language: en
-    speaker: en_103
-    silero_speaker: v3_en
-  en_104:
-    language: en
-    speaker: en_104
-    silero_speaker: v3_en
-  en_105:
-    language: en
-    speaker: en_105
-    silero_speaker: v3_en
-  en_106:
-    language: en
-    speaker: en_106
-    silero_speaker: v3_en
-  en_107:
-    language: en
-    speaker: en_107
-    silero_speaker: v3_en
-  en_108:
-    language: en
-    speaker: en_108
-    silero_speaker: v3_en
-  en_109:
-    language: en
-    speaker: en_109
-    silero_speaker: v3_en
-  en_110:
-    language: en
-    speaker: en_110
-    silero_speaker: v3_en
-  en_111:
-    language: en
-    speaker: en_111
-    silero_speaker: v3_en
-  en_112:
-    language: en
-    speaker: en_112
-    silero_speaker: v3_en
-  en_113:
-    language: en
-    speaker: en_113
-    silero_speaker: v3_en
-  en_114:
-    language: en
-    speaker: en_114
-    silero_speaker: v3_en
-  en_115:
-    language: en
-    speaker: en_115
-    silero_speaker: v3_en
-  en_116:
-    language: en
-    speaker: en_116
-    silero_speaker: v3_en
-  en_117:
-    language: en
-    speaker: en_117
-    silero_speaker: v3_en
-  random:
-    language: en
-    speaker: random
-    silero_speaker: v3_en
-  # Russian voices (v3_ru/ru_v3 model) - use ru_v3 for better compatibility
-  ru_aidar:
-    language: ru
-    speaker: aidar
-    silero_speaker: ru_v3
-  ru_baya:
-    language: ru
-    speaker: baya
-    silero_speaker: ru_v3
-  ru_kseniya:
-    language: ru
-    speaker: kseniya
-    silero_speaker: ru_v3
-  ru_xenia:
-    language: ru
-    speaker: xenia
-    silero_speaker: ru_v3
-  ru_eugene:
-    language: ru
-    speaker: eugene
-    silero_speaker: ru_v3
-  ru_random:
-    language: ru
-    speaker: random
-    silero_speaker: ru_v3
-  # German voices (v3_de model)
-  de_bernd_ungerer:
-    language: de
-    speaker: bernd_ungerer
-    silero_speaker: v3_de
-  de_eva_k:
-    language: de
-    speaker: eva_k
-    silero_speaker: v3_de
-  de_friedrich:
-    language: de
-    speaker: friedrich
-    silero_speaker: v3_de
-  de_hokuspokus:
-    language: de
-    speaker: hokuspokus
-    silero_speaker: v3_de
-  de_karlsson:
-    language: de
-    speaker: karlsson
-    silero_speaker: v3_de
-  de_random:
-    language: de
-    speaker: random
-    silero_speaker: v3_de
-  # Spanish voices (v3_es model)
-  es_0:
-    language: es
-    speaker: es_0
-    silero_speaker: v3_es
-  es_1:
-    language: es
-    speaker: es_1
-    silero_speaker: v3_es
-  es_2:
-    language: es
-    speaker: es_2
-    silero_speaker: v3_es
-  es_random:
-    language: es
-    speaker: random
-    silero_speaker: v3_es
-  # French voices (v3_fr model)
-  fr_0:
-    language: fr
-    speaker: fr_0
-    silero_speaker: v3_fr
-  fr_1:
-    language: fr
-    speaker: fr_1
-    silero_speaker: v3_fr
-  fr_2:
-    language: fr
-    speaker: fr_2
-    silero_speaker: v3_fr
-  fr_3:
-    language: fr
-    speaker: fr_3
-    silero_speaker: v3_fr
-  fr_4:
-    language: fr
-    speaker: fr_4
-    silero_speaker: v3_fr
-  fr_5:
-    language: fr
-    speaker: fr_5
-    silero_speaker: v3_fr
-  fr_random:
-    language: fr
-    speaker: random
-    silero_speaker: v3_fr
-tts-1-kokoro:
-  # OpenAI-compatible voice aliases (Kokoro's intentional OpenAI-themed voices)
-  alloy:
-    lang_code: a
-    kokoro_voice: af_alloy
-  echo:
-    lang_code: a
-    kokoro_voice: am_echo
-  fable:
-    lang_code: b
-    kokoro_voice: bm_fable
-  onyx:
-    lang_code: a
-    kokoro_voice: am_onyx
-  nova:
-    lang_code: a
-    kokoro_voice: af_nova
-  shimmer:
-    lang_code: a
-    kokoro_voice: af_sky
-  # Female American voices
-  af_heart:
-    lang_code: a
-    kokoro_voice: af_heart
-  af_bella:
-    lang_code: a
-    kokoro_voice: af_bella
-  af_nicole:
-    lang_code: a
-    kokoro_voice: af_nicole
-  af_aoede:
-    lang_code: a
-    kokoro_voice: af_aoede
-  af_kore:
-    lang_code: a
-    kokoro_voice: af_kore
-  af_sarah:
-    lang_code: a
-    kokoro_voice: af_sarah
-  af_nova:
-    lang_code: a
-    kokoro_voice: af_nova
-  af_sky:
-    lang_code: a
-    kokoro_voice: af_sky
-  af_alloy:
-    lang_code: a
-    kokoro_voice: af_alloy
-  af_jessica:
-    lang_code: a
-    kokoro_voice: af_jessica
-  af_river:
-    lang_code: a
-    kokoro_voice: af_river
-  # Male American voices
-  am_michael:
-    lang_code: a
-    kokoro_voice: am_michael
-  am_fenrir:
-    lang_code: a
-    kokoro_voice: am_fenrir
-  am_puck:
-    lang_code: a
-    kokoro_voice: am_puck
-  am_echo:
-    lang_code: a
-    kokoro_voice: am_echo
-  am_eric:
-    lang_code: a
-    kokoro_voice: am_eric
-  am_liam:
-    lang_code: a
-    kokoro_voice: am_liam
-  am_onyx:
-    lang_code: a
-    kokoro_voice: am_onyx
-  am_santa:
-    lang_code: a
-    kokoro_voice: am_santa
-  am_adam:
-    lang_code: a
-    kokoro_voice: am_adam
-  # Female British voices
-  bf_emma:
-    lang_code: b
-    kokoro_voice: bf_emma
-  bf_isabella:
-    lang_code: b
-    kokoro_voice: bf_isabella
-  bf_alice:
-    lang_code: b
-    kokoro_voice: bf_alice
-  bf_lily:
-    lang_code: b
-    kokoro_voice: bf_lily
-  # Male British voices
-  bm_george:
-    lang_code: b
-    kokoro_voice: bm_george
-  bm_fable:
-    lang_code: b
-    kokoro_voice: bm_fable
-  bm_lewis:
-    lang_code: b
-    kokoro_voice: bm_lewis
-  bm_daniel:
-    lang_code: b
-    kokoro_voice: bm_daniel
+    ref_audio: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav
+    ref_text: "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it!"
+    language: English
+
+  # Custom voice example - add your own reference audio
+  # my_voice:
+  #   ref_audio: voices/my_voice_sample.wav
+  #   ref_text: "The exact text spoken in the reference audio file"
+  #   language: English
+
+# Other models are disabled by default. Uncomment to enable.
+# See the git history for full voice configurations.
+
+# tts-1:
+#   # Piper TTS voices (fast CPU inference)
+#   alloy:
+#     model: /app/voices/en/en_US/libritts_r/medium/en_US-libritts_r-medium.onnx
+#     speaker: 79
+
+# tts-1-hd:
+#   # XTTS voice cloning
+#   alloy:
+#     model: xtts
+#     speaker: voices/alloy.wav
+
+# tts-1-silero:
+#   # Silero TTS (CPU-friendly)
+#   en_0:
+#     language: en
+#     speaker: en_0
+#     silero_speaker: v3_en
+
+# tts-1-kokoro:
+#   # Kokoro TTS (lightweight 82M params)
+#   alloy:
+#     lang_code: a
+#     kokoro_voice: af_alloy