feat: integrate Silero TTS as default TTS engine

- Add SileroTTS service with GPU/CPU support
- Update TTS API to support configurable engines (silero/edge)
- Add TTS_ENGINE and SILERO_SPEAKER config options
- Set Silero as default TTS engine
- Update requirements.txt with silero dependency

app/services/silero_tts_service.py

@@ -0,0 +1,87 @@
+"""Silero TTS service for text-to-speech conversion."""
+import os
+import logging
+from typing import Optional
+import torch
+import numpy as np
+import soundfile as sf
+
+logger = logging.getLogger(__name__)
+
+
+class SileroTTSService:
+    """Service for converting text to speech using Silero TTS."""
+    
+    def __init__(self, device: str = 'cuda', speaker: str = 'v5_ru'):
+        """
+        Initialize Silero TTS service.
+        
+        Args:
+            device: Device to use ('cuda' or 'cpu')
+            speaker: Speaker model to use (e.g., 'v5_ru')
+        """
+        self.device = device if torch.cuda.is_available() else 'cpu'
+        self.speaker = speaker
+        self.model = None
+        self._load_model()
+        logger.info(f"Silero TTS Service initialized with speaker: {speaker}, device: {self.device}")
+    
+    def _load_model(self):
+        """Load Silero TTS model."""
+        try:
+            # Import silero here to avoid loading at module level
+            from silero import silero_tts
+            self.model, _ = silero_tts(
+                language='ru',
+                speaker=self.speaker,
+                device=self.device
+            )
+            logger.info(f"Silero TTS model loaded successfully on {self.device}")
+        except Exception as e:
+            logger.error(f"Failed to load Silero TTS model: {e}")
+            raise
+    
+    def text_to_speech(self, text: str, output_path: str, speed: float = 1.0) -> str:
+        """
+        Convert text to speech and save to file.
+        
+        Args:
+            text: Text to convert
+            output_path: Path to save audio file
+            speed: Speech speed multiplier (not supported by Silero, kept for API compatibility)
+        
+        Returns:
+            Path to generated audio file
+        """
+        logger.info(f"Generating TTS with Silero for text: {text[:50]}...")
+        
+        try:
+            # Generate audio
+            audio = self.model.apply_tts(text=text)
+            
+            # Convert to numpy array
+            audio_np = audio.numpy() if torch.is_tensor(audio) else np.array(audio)
+            
+            # Ensure float32 format
+            audio_np = audio_np.astype(np.float32)
+            
+            # Save as WAV (Silero uses 16000 Hz by default)
+            sf.write(output_path, audio_np, 16000)
+            
+            logger.info(f"Silero TTS generated: {output_path}")
+            return output_path
+            
+        except Exception as e:
+            logger.error(f"Silero TTS generation failed: {e}")
+            raise
+    
+    def get_supported_voices(self) -> list:
+        """Get list of supported speakers."""
+        # Silero Russian models
+        voices = [
+            "v5_ru",  # Default Russian voice (best quality)
+            "v4_ru",
+            "v3_1_ru",
+            "v3_ru",
+        ]
+        return voices