fix: Update Piper TTS to support voice preview and improve setup

- Fixed preview endpoint to work with both Kokoro and Piper engines
- Improved voice model download with curl fallback and SSL handling
- Added better error messages when models are missing
- Created PIPER_SETUP.md with manual download instructions
- Added test_piper_direct.py for testing Piper installation
- Updated README with voice model download requirements

Note: Piper voice models require manual download from GitHub releases
due to HuggingFace URL structure. See PIPER_SETUP.md for instructions.

backend/services/piper_tts_service.py

@@ -72,20 +72,64 @@ class PiperTTSService:
         
         # Check if model already exists
         if os.path.exists(model_path) and os.path.exists(config_path):
+            logger.info(f"Voice model {voice} already exists")
             return True
         
         logger.info(f"Downloading Piper voice model: {voice}")
+        logger.info(f"Models directory: {self.models_dir}")
         
-        # Download using piper's built-in download
+        # Try to download using wget/curl as Piper's download might not work
         try:
-            subprocess.run(
-                ["piper", "--model", voice, "--download-dir", self.models_dir],
-                check=True,
-                timeout=60
-            )
-            return os.path.exists(model_path)
-        except subprocess.SubprocessError as e:
+            # Piper models are hosted on HuggingFace
+            base_url = "https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0"
+            
+            # Determine language prefix from voice ID
+            lang_prefix = voice.split('-')[0]  # e.g., "en_US" from "en_US-lessac-medium"
+            
+            model_url = f"{base_url}/{lang_prefix}/{voice}.onnx"
+            config_url = f"{base_url}/{lang_prefix}/{voice}.onnx.json"
+            
+            logger.info(f"Downloading from: {model_url}")
+            
+            # Try using curl/wget first (more reliable)
+            try:
+                # Try curl
+                subprocess.run(
+                    ["curl", "-L", "-o", model_path, model_url],
+                    check=True,
+                    timeout=120,
+                    capture_output=True
+                )
+                subprocess.run(
+                    ["curl", "-L", "-o", config_path, config_url],
+                    check=True,
+                    timeout=120,
+                    capture_output=True
+                )
+            except (subprocess.SubprocessError, FileNotFoundError):
+                # Fall back to urllib with SSL context
+                import urllib.request
+                import ssl
+                
+                # Create SSL context that doesn't verify certificates (for development)
+                ssl_context = ssl.create_default_context()
+                ssl_context.check_hostname = False
+                ssl_context.verify_mode = ssl.CERT_NONE
+                
+                urllib.request.urlretrieve(model_url, model_path, context=ssl_context)
+                urllib.request.urlretrieve(config_url, config_path, context=ssl_context)
+            
+            if os.path.exists(model_path) and os.path.exists(config_path):
+                logger.info(f"Successfully downloaded voice model: {voice}")
+                return True
+            else:
+                logger.error(f"Download completed but files not found")
+                return False
+                
+        except Exception as e:
             logger.error(f"Failed to download voice model {voice}: {e}")
+            logger.error(f"Please manually download from: https://github.com/rhasspy/piper/releases/tag/v1.2.0")
+            logger.error(f"See backend/PIPER_SETUP.md for detailed instructions")
             return False
     
     def synthesize_with_pitch(
@@ -109,14 +153,25 @@ class PiperTTSService:
         """
         # Ensure voice model is available
         if not self._download_voice_model(voice):
-            logger.warning(f"Voice {voice} not available, using default")
-            voice = self.DEFAULT_VOICE
-            self._download_voice_model(voice)
+            error_msg = (
+                f"Piper voice model '{voice}' not found. "
+                f"Please download voice models manually. "
+                f"See backend/PIPER_SETUP.md for instructions."
+            )
+            logger.error(error_msg)
+            raise FileNotFoundError(error_msg)
         
         model_path = os.path.join(self.models_dir, f"{voice}.onnx")
         
+        # Check if model file exists
+        if not os.path.exists(model_path):
+            logger.error(f"Model file not found: {model_path}")
+            return np.array([], dtype=np.float32)
+        
         # Generate audio with Piper
         try:
+            logger.info(f"Running Piper with model: {model_path}")
+            
             # Run piper command
             process = subprocess.Popen(
                 ["piper", "--model", model_path, "--output-raw"],
@@ -129,7 +184,8 @@ class PiperTTSService:
             stdout, stderr = process.communicate(input=text.encode('utf-8'), timeout=30)
             
             if process.returncode != 0:
-                logger.error(f"Piper synthesis failed: {stderr.decode()}")
+                error_msg = stderr.decode() if stderr else "Unknown error"
+                logger.error(f"Piper synthesis failed (exit code {process.returncode}): {error_msg}")
                 return np.array([], dtype=np.float32)
             
             # Convert raw audio to numpy array