utils/tts.py

import os
import logging
import time
import soundfile as sf

logger = logging.getLogger(__name__)

# Wrap the problematic import in a try-except block
try:
    from kokoro import KPipeline
    KOKORO_AVAILABLE = True
except AttributeError as e:
    # Specifically catch the EspeakWrapper.set_data_path error
    if "EspeakWrapper" in str(e) and "set_data_path" in str(e):
        logger.warning("Kokoro import failed due to EspeakWrapper.set_data_path issue")
        KOKORO_AVAILABLE = False
    else:
        # Re-raise if it's a different error
        raise

class TTSEngine:
    def __init__(self, lang_code='z'):
        """Initialize TTS Engine with Kokoro
        
        Args:
            lang_code (str): Language code ('a' for US English, 'b' for British English,
                           'j' for Japanese, 'z' for Mandarin Chinese)
        """
        logger.info("Initializing TTS Engine")
        if not KOKORO_AVAILABLE:
            logger.warning("Using dummy TTS implementation as Kokoro is not available")
            self.pipeline = None
        else:
            self.pipeline = KPipeline(lang_code=lang_code)
            logger.info("TTS engine initialized with Kokoro")

    def generate_speech(self, text: str, voice: str = 'af_heart', speed: float = 1.0) -> str:
        """Generate speech from text using Kokoro
        
        Args:
            text (str): Input text to synthesize
            voice (str): Voice ID to use (e.g., 'af_heart', 'af_bella', etc.)
            speed (float): Speech speed multiplier (0.5 to 2.0)
            
        Returns:
            str: Path to the generated audio file
        """
        logger.info(f"Generating speech for text length: {len(text)}")
        
        try:
            # Create output directory if it doesn't exist
            os.makedirs("temp/outputs", exist_ok=True)
            
            # Generate unique output path
            output_path = f"temp/outputs/output_{int(time.time())}.wav"
            
            if not KOKORO_AVAILABLE:
                # Generate a simple sine wave as dummy audio
                import numpy as np
                sample_rate = 24000
                duration = 3.0  # seconds
                t = np.linspace(0, duration, int(sample_rate * duration), False)
                tone = np.sin(2 * np.pi * 440 * t) * 0.3
                
                logger.info(f"Saving dummy audio to {output_path}")
                sf.write(output_path, tone, sample_rate)
                logger.info(f"Dummy audio generation complete: {output_path}")
                return output_path
            
            # Get the first generated segment
            # We only take the first segment since the original code handled single segments
            generator = self.pipeline(text, voice=voice, speed=speed)
            for _, _, audio in generator:
                logger.info(f"Saving audio to {output_path}")
                sf.write(output_path, audio, 24000)
                break
            
            logger.info(f"Audio generation complete: {output_path}")
            return output_path

        except Exception as e:
            logger.error(f"TTS generation failed: {str(e)}", exc_info=True)
            raise

    def generate_speech_stream(self, text: str, voice: str = 'af_heart', speed: float = 1.0):
        """Generate speech from text and yield each segment
        
        Args:
            text (str): Input text to synthesize
            voice (str): Voice ID to use (e.g., 'af_heart', 'af_bella', etc.)
            speed (float): Speech speed multiplier (0.5 to 2.0)
            
        Yields:
            tuple: (sample_rate, audio_data) pairs for each segment
        """
        try:
            if not KOKORO_AVAILABLE:
                # Generate dummy audio chunks
                import numpy as np
                sample_rate = 24000
                duration = 1.0  # seconds per chunk
                
                # Create 3 chunks of dummy audio
                for i in range(3):
                    t = np.linspace(0, duration, int(sample_rate * duration), False)
                    freq = 440 + (i * 220)  # Different frequency for each chunk
                    tone = np.sin(2 * np.pi * freq * t) * 0.3
                    yield sample_rate, tone
                return
                
            generator = self.pipeline(text, voice=voice, speed=speed)
            for _, _, audio in generator:
                yield 24000, audio
                
        except Exception as e:
            logger.error(f"TTS streaming failed: {str(e)}", exc_info=True)
            raise

# Initialize TTS engine with cache decorator if using Streamlit
def get_tts_engine(lang_code='a'):
    """Get or create TTS engine instance
    
    Args:
        lang_code (str): Language code for the pipeline
        
    Returns:
        TTSEngine: Initialized TTS engine instance
    """
    try:
        import streamlit as st
        @st.cache_resource
        def _get_engine():
            return TTSEngine(lang_code)
        return _get_engine()
    except ImportError:
        return TTSEngine(lang_code)

def generate_speech(text: str, voice: str = 'af_heart', speed: float = 1.0) -> str:
    """Public interface for TTS generation
    
    Args:
        text (str): Input text to synthesize
        voice (str): Voice ID to use
        speed (float): Speech speed multiplier
        
    Returns:
        str: Path to generated audio file
    """
    engine = get_tts_engine()
    return engine.generate_speech(text, voice, speed)