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

.env.example

@@ -9,9 +9,15 @@ OUTPUT_DIR=./output
 SEED_VC_MODEL=seed-uvit-whisper-small-wavenet
 SEED_VC_REPO_PATH=./Seed-VC
 
+# TTS Engine Configuration
+TTS_ENGINE=silero  # Options: silero, edge
+
 # Edge-TTS Configuration
 EDGE_TTS_VOICE=ru-RU-SvetlanaNeural
 
+# Silero TTS Configuration
+SILERO_SPEAKER=v5_ru
+
 # Flask Configuration
 FLASK_PORT=5000
 FLASK_DEBUG=false

app/api/tts.py

@@ -7,6 +7,7 @@ from flask import current_app, request, send_file
 from flask_restx import Namespace, Resource, fields
 from app.models.voice import Voice
 from app.services.edge_tts_service import EdgeTTSService
+from app.services.silero_tts_service import SileroTTSService
 from app.services.seed_vc_service import SeedVCService
 
 logger = logging.getLogger(__name__)
@@ -110,10 +111,17 @@ class TTS(Resource):
         final_output = os.path.join(output_dir, f"{output_id}_converted.wav")
         
         try:
-            # Step 1: Generate TTS with Edge-TTS
-            logger.info("Step 1: Generating TTS with Edge-TTS")
-            edge_tts_voice = current_app.config['EDGE_TTS_VOICE']
-            tts_service = EdgeTTSService(voice=edge_tts_voice)
+            # Step 1: Generate TTS
+            tts_engine = current_app.config.get('TTS_ENGINE', 'silero')
+            logger.info(f"Step 1: Generating TTS with {tts_engine}")
+            
+            if tts_engine == 'silero':
+                silero_speaker = current_app.config.get('SILERO_SPEAKER', 'v5_ru')
+                tts_service = SileroTTSService(speaker=silero_speaker)
+            else:
+                edge_tts_voice = current_app.config['EDGE_TTS_VOICE']
+                tts_service = EdgeTTSService(voice=edge_tts_voice)
+            
             tts_service.text_to_speech(text, tts_output, speed=speed)
             
             # Step 2: Convert voice with Seed-VC

app/config.py

@@ -18,9 +18,15 @@ class Config:
     SEED_VC_MODEL = os.getenv('SEED_VC_MODEL', 'seed-uvit-whisper-small-wavenet')
     SEED_VC_REPO_PATH = os.getenv('SEED_VC_REPO_PATH', './Seed-VC')
     
+    # TTS Engine
+    TTS_ENGINE = os.getenv('TTS_ENGINE', 'silero')  # Default to Silero
+    
     # Edge-TTS
     EDGE_TTS_VOICE = os.getenv('EDGE_TTS_VOICE', 'ru-RU-SvetlanaNeural')
     
+    # Silero TTS
+    SILERO_SPEAKER = os.getenv('SILERO_SPEAKER', 'v5_ru')
+    
     # Flask
     FLASK_PORT = int(os.getenv('FLASK_PORT', '5000'))
     FLASK_DEBUG = os.getenv('FLASK_DEBUG', 'false').lower() == 'true'

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

requirements.txt

@@ -2,6 +2,7 @@ flask>=2.3.0
 flask-restx>=1.3.0
 python-dotenv>=1.0.0
 edge-tts>=6.1.0
+silero>=0.5.0
 torch>=2.0.0
 torchaudio>=2.0.0
 numpy>=1.24.0