Skip to content

anbeime/skill / skills/tts-voice-synthesis

tts-voice-synthesis

Verified

影片与视频编辑、内容创作者在制作视频配音或有声书时,当需要克隆音色、生成情感化配音或流式实时语音合成请用此技能。支持1.7B高质量与0.6B快速双模型,一键实现多语言方言配音,让语音创作更高效更自然。

FileKindSize
SKILL.mdskill md5.2 KB
references/emotion_guide.mdmarkdown7.1 KB
references/model_config.mdmarkdown5.4 KB
references/usage_examples.mdmarkdown10.1 KB
scripts/tts_generate.pyscript12.4 KB
scripts/voice_clone.pyscript11.1 KB

scripts/tts_generate.py

View on GitHub
#!/usr/bin/env python3
"""
TTS Voice Generation Script

This script provides a unified interface for text-to-speech generation
supporting multiple TTS models (Fish-Speech, ChatTTS, CosyVoice).
"""

import argparse
import logging
import os
import sys
import warnings
warnings.filterwarnings('ignore')

import torch
import torchaudio
import numpy as np
from pathlib import Path

# Configure logging
logging.basicConfig(
    level=logging.INFO,
    format='[%(asctime)s] %(levelname)s: %(message)s'
)
logger = logging.getLogger(__name__)


def parse_args():
    """Parse command line arguments"""
    parser = argparse.ArgumentParser(
        description="Generate speech from text using TTS models"
    )

    # Input arguments
    parser.add_argument("--text", type=str, default=None,
                      help="Input text to synthesize")
    parser.add_argument("--text_file", type=str, default=None,
                      help="Input text file path")
    parser.add_argument("--output_path", type=str, required=True,
                      help="Output audio file path")

    # Model selection
    parser.add_argument("--model_name", type=str, default="fish-speech-1.5",
                      choices=["fish-speech-1.5", "chattts", "cosyvoice"],
                      help="TTS model to use")
    parser.add_argument("--model_size", type=str, default=None,
                      help="Model size (1.7B, 0.6B)")
    parser.add_argument("--model_path", type=str, default=None,
                      help="Path to model weights")

    # Voice selection
    parser.add_argument("--voice", type=str, default="default_female",
                      help="Voice name or path")
    parser.add_argument("--voice_path", type=str, default=None,
                      help="Path to custom voice model")

    # Audio parameters
    parser.add_argument("--sample_rate", type=int, default=24000,
                      help="Audio sample rate")
    parser.add_argument("--speed", type=float, default=1.0,
                      help="Speech speed (0.5-2.0)")
    parser.add_argument("--pitch", type=float, default=1.0,
                      help="Pitch adjustment (0.5-2.0)")
    parser.add_argument("--volume", type=float, default=1.0,
                      help="Volume adjustment (0.5-1.5)")
    parser.add_argument("--audio_format", type=str, default="wav",
                      choices=["wav", "mp3", "flac"],
                      help="Output audio format")

    # Emotion parameters
    parser.add_argument("--emotion", type=str, default="neutral",
                      choices=["neutral", "happy", "sad", "angry",
                              "excited", "calm", "curious", "serious"],
                      help="Emotion type")
    parser.add_argument("--emotion_strength", type=float, default=0.8,
                      help="Emotion strength (0.0-1.0)")
    parser.add_argument("--emotion_per_sentence", type=bool, default=False,
                      help="Analyze emotion per sentence")

    # Streaming parameters
    parser.add_argument("--streaming", action="store_true", default=False,
                      help="Enable streaming generation")
    parser.add_argument("--chunk_size", type=int, default=128,
                      help="Chunk size for streaming")

    # Model-specific parameters
    parser.add_argument("--temperature", type=float, default=0.7,
                      help="Sampling temperature (0.1-1.0)")
    parser.add_argument("--top_p", type=float, default=0.9,
                      help="Top-p sampling probability")
    parser.add_argument("--top_k", type=int, default=20,
                      help="Top-k sampling")

    # Device
    parser.add_argument("--device", type=str, default=None,
                      help="Device to use (cuda, cpu)")

    return parser.parse_args()


class TTSGenerator:
    """Unified TTS Generator class"""

    def __init__(self, model_name, model_path=None, device=None):
        """
        Initialize TTS Generator

        Args:
            model_name: Name of the TTS model
            model_path: Path to model weights
            device: Device to use (cuda, cpu)
        """
        self.model_name = model_name
        self.device = device or ("cuda" if torch.cuda.is_available() else "cpu")
        self.model = None
        self.model_loaded = False

        logger.info(f"Initializing TTS Generator with model: {model_name}")
        logger.info(f"Device: {self.device}")

        # Load model
        self._load_model(model_path)

    def _load_model(self, model_path):
        """Load the TTS model"""
        logger.info(f"Loading model...")

        try:
            if self.model_name == "fish-speech-1.5":
                self._load_fish_speech(model_path)
            elif self.model_name == "chattts":
                self._load_chattts(model_path)
            elif self.model_name == "cosyvoice":
                self._load_cosyvoice(model_path)
            else:
                raise ValueError(f"Unsupported model: {self.model_name}")

            self.model_loaded = True
            logger.info("Model loaded successfully")

        except Exception as e:
            logger.error(f"Failed to load model: {e}")
            logger.warning("Model will be loaded on-the-fly")
            self.model_loaded = False

    def _load_fish_speech(self, model_path):
        """Load Fish-Speech model"""
        # Placeholder for Fish-Speech model loading
        logger.info("This is a placeholder for Fish-Speech model loading")
        logger.info("In production, this would load the actual Fish-Speech model")
        self.model = "fish-speech-placeholder"

    def _load_chattts(self, model_path):
        """Load ChatTTS model"""
        # Placeholder for ChatTTS model loading
        logger.info("This is a placeholder for ChatTTS model loading")
        logger.info("In production, this would load the actual ChatTTS model")
        self.model = "chattts-placeholder"

    def _load_cosyvoice(self, model_path):
        """Load CosyVoice model"""
        # Placeholder for CosyVoice model loading
        logger.info("This is a placeholder for CosyVoice model loading")
        logger.info("In production, this would load the actual CosyVoice model")
        self.model = "cosyvoice-placeholder"

    def synthesize(self, text, voice="default_female", speed=1.0, pitch=1.0,
                   volume=1.0, emotion="neutral", emotion_strength=0.8,
                   streaming=False, **kwargs):
        """
        Synthesize speech from text

        Args:
            text: Input text
            voice: Voice name or path
            speed: Speech speed
            pitch: Pitch adjustment
            volume: Volume adjustment
            emotion: Emotion type
            emotion_strength: Emotion strength
            streaming: Enable streaming

        Returns:
            Audio data (numpy array)
        """
        if not text:
            raise ValueError("Text cannot be empty")

        logger.info(f"Synthesizing speech for: {text[:50]}...")
        logger.info(f"Voice: {voice}, Speed: {speed}, Pitch: {pitch}")
        logger.info(f"Emotion: {emotion}, Strength: {emotion_strength}")

        # Apply emotion adjustments
        speed, pitch, volume = self._apply_emotion(
            emotion, emotion_strength, speed, pitch, volume
        )

        # Synthesize (placeholder implementation)
        audio = self._synthesize_placeholder(
            text, voice, speed, pitch, volume
        )

        logger.info("Speech synthesis completed")
        return audio

    def _apply_emotion(self, emotion, strength, speed, pitch, volume):
        """Apply emotion adjustments to parameters"""
        # Emotion parameter mappings
        emotion_params = {
            "neutral": {"speed": 1.0, "pitch": 1.0, "volume": 1.0},
            "happy": {"speed": 1.2, "pitch": 1.1, "volume": 1.1},
            "sad": {"speed": 0.85, "pitch": 0.85, "volume": 0.85},
            "angry": {"speed": 1.25, "pitch": 1.25, "volume": 1.25},
            "excited": {"speed": 1.4, "pitch": 1.3, "volume": 1.3},
            "calm": {"speed": 0.95, "pitch": 1.0, "volume": 0.95},
            "curious": {"speed": 0.95, "pitch": 1.05, "volume": 1.0},
            "serious": {"speed": 1.0, "pitch": 1.0, "volume": 1.0},
        }

        # Get base emotion parameters
        params = emotion_params.get(emotion, emotion_params["neutral"])

        # Apply strength
        speed = speed * (1 + (params["speed"] - 1) * strength)
        pitch = pitch * (1 + (params["pitch"] - 1) * strength)
        volume = volume * (1 + (params["volume"] - 1) * strength)

        # Clamp values
        speed = np.clip(speed, 0.5, 2.0)
        pitch = np.clip(pitch, 0.5, 2.0)
        volume = np.clip(volume, 0.5, 1.5)

        logger.debug(f"Adjusted parameters - Speed: {speed:.2f}, "
                    f"Pitch: {pitch:.2f}, Volume: {volume:.2f}")

        return speed, pitch, volume

    def _synthesize_placeholder(self, text, voice, speed, pitch, volume):
        """
        Placeholder synthesis implementation

        In production, this would call the actual TTS model
        """
        logger.warning("=" * 80)
        logger.warning("This is a placeholder implementation!")
        logger.warning("In production, this would:")
        logger.warning(f"  1. Load the {self.model_name} model")
        logger.warning(f"  2. Process text: '{text[:50]}...'")
        logger.warning(f"  3. Use voice: {voice}")
        logger.warning(f"  4. Apply speed={speed:.2f}, pitch={pitch:.2f}, volume={volume:.2f}")
        logger.warning(f"  5. Generate audio waveform")
        logger.warning("=" * 80)

        # Generate dummy audio for demonstration
        # In production, this would be the actual audio from the TTS model
        duration = len(text) * 0.1 / speed  # Estimate duration
        sample_rate = 24000
        num_samples = int(duration * sample_rate)

        # Generate silent audio (placeholder)
        audio = np.zeros(num_samples, dtype=np.float32)

        return audio

    def save_audio(self, audio, output_path, sample_rate=24000,
                  audio_format="wav", volume=1.0):
        """
        Save audio to file

        Args:
            audio: Audio data (numpy array)
            output_path: Output file path
            sample_rate: Sample rate
            audio_format: Audio format (wav, mp3, flac)
            volume: Volume adjustment
        """
        # Create output directory
        os.makedirs(os.path.dirname(output_path), exist_ok=True)

        # Apply volume
        audio = audio * volume

        # Clamp values
        audio = np.clip(audio, -1.0, 1.0)

        # Convert to tensor
        audio_tensor = torch.from_numpy(audio).unsqueeze(0)

        # Save audio
        if audio_format == "wav":
            torchaudio.save(output_path, audio_tensor, sample_rate)
        elif audio_format == "mp3":
            # For MP3, you would need additional library (like pydub or ffmpeg)
            logger.warning("MP3 format requires additional libraries")
            # Save as WAV as fallback
            wav_path = output_path.replace(".mp3", ".wav")
            torchaudio.save(wav_path, audio_tensor, sample_rate)
            logger.info(f"Saved as WAV (MP3 not supported): {wav_path}")
        elif audio_format == "flac":
            torchaudio.save(output_path, audio_tensor, sample_rate,
                          format="flac")

        logger.info(f"Audio saved to: {output_path}")


def main():
    """Main function"""
    args = parse_args()

    # Read input text
    if args.text:
        text = args.text
    elif args.text_file:
        with open(args.text_file, 'r', encoding='utf-8') as f:
            text = f.read()
    else:
        logger.error("Either --text or --text_file must be provided")
        return 1

    # Initialize TTS generator
    generator = TTSGenerator(
        model_name=args.model_name,
        model_path=args.model_path,
        device=args.device
    )

    # Synthesize speech
    audio = generator.synthesize(
        text=text,
        voice=args.voice,
        speed=args.speed,
        pitch=args.pitch,
        volume=args.volume,
        emotion=args.emotion,
        emotion_strength=args.emotion_strength,
        streaming=args.streaming
    )

    # Save audio
    generator.save_audio(
        audio=audio,
        output_path=args.output_path,
        sample_rate=args.sample_rate,
        audio_format=args.audio_format,
        volume=1.0  # Volume already applied in synthesis
    )

    logger.info("Done!")
    return 0


if __name__ == "__main__":
    sys.exit(main())
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall