cli.py

import argparse
import sys
import os
from voice_clone import VoiceCloner

def main():
    parser = argparse.ArgumentParser(description="Voice Clone CLI")
    subparsers = parser.add_subparsers(dest="command", help="Available commands")

    # Clone command
    clone_parser = subparsers.add_parser("clone", help="Clone voice from reference audio")
    clone_parser.add_argument("--ref-audio", required=True, help="Path to reference audio file")
    clone_parser.add_argument("--ref-text", required=True, help="Transcript of reference audio")
    clone_parser.add_argument("--text", required=True, help="Target text to synthesize")
    clone_parser.add_argument("--model", default="1.7B", choices=["1.7B", "0.6B"], help="Model size (default: 1.7B)")
    clone_parser.add_argument("--output", default="output.wav", help="Output file path (default: output.wav)")

    # Save Profile command
    save_parser = subparsers.add_parser("save-profile", help="Save reference as a profile")
    save_parser.add_argument("--name", required=True, help="Profile name")
    save_parser.add_argument("--ref-audio", required=True, help="Path to reference audio file")
    save_parser.add_argument("--ref-text", required=True, help="Transcript of reference audio")

    # List Profiles command
    subparsers.add_parser("list-profiles", help="List available profiles")

    # Use Profile command
    use_parser = subparsers.add_parser("use-profile", help="Generate audio using a saved profile")
    use_parser.add_argument("--name", required=True, help="Profile name")
    use_parser.add_argument("--text", required=True, help="Target text to synthesize")
    use_parser.add_argument("--model", default="1.7B", choices=["1.7B", "0.6B"], help="Model size (default: 1.7B)")
    use_parser.add_argument("--output", default="output.wav", help="Output file path (default: output.wav)")

    # Delete Profile command
    delete_parser = subparsers.add_parser("delete-profile", help="Delete a saved profile")
    delete_parser.add_argument("--name", required=True, help="Profile name to delete")

    args = parser.parse_args()

    if not args.command:
        parser.print_help()
        return

    # Helper to load model
    def get_cloner(model_choice):
        model_path = "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16"
        if model_choice == "0.6B":
            model_path = "mlx-community/Qwen3-TTS-12Hz-0.6B-Base-bf16"
        
        try:
            return VoiceCloner(model_path=model_path)
        except Exception as e:
            print(f"Error initializing model: {e}")
            sys.exit(1)

    if args.command == "clone":
        cloner = get_cloner(args.model)
        try:
            output = cloner.clone_and_generate(args.ref_audio, args.ref_text, args.text)
            if args.output != "output.wav":
                os.rename(output, args.output)
                output = args.output
            print(f"Successfully generated audio at: {output}")
        except Exception as e:
            print(f"Error cloning voice: {e}")

    elif args.command == "save-profile":
        # No need to load model just to save profile, but VoiceCloner handles paths.
        # We can instantiate with default but not load if we refactor, but current init loads.
        # Let's instantiate and unload quickly or just use it.
        # Ideally VoiceCloner shouldn't load model in init if we only want utility,
        # but for now we follow existing pattern.
        cloner = get_cloner("1.7B") # Model irrelevant for saving
        try:
            msg = cloner.save_profile(args.name, args.ref_audio, args.ref_text)
            print(msg)
        except Exception as e:
            print(f"Error saving profile: {e}")

    elif args.command == "list-profiles":
        # Hack: instantiating just to list might be slow due to load. 
        # Refactoring VoiceCloner to lazy load would be better, but avoiding big refactors for now.
        # Actually initializing VoiceCloner loads the model. 
        # Let's just manually check the directory for speed.
        if os.path.exists("profiles"):
             profiles = [d for d in os.listdir("profiles") if os.path.isdir(os.path.join("profiles", d))]
             if profiles:
                 print("Available Profiles:")
                 for p in profiles:
                     print(f" - {p}")
             else:
                 print("No profiles found.")
        else:
            print("No profiles directory found.")

    elif args.command == "use-profile":
        cloner = get_cloner(args.model)
        try:
            audio_path, ref_text = cloner.load_profile(args.name)
            output = cloner.clone_and_generate(audio_path, ref_text, args.text)
            if args.output != "output.wav":
                os.rename(output, args.output)
                output = args.output
            print(f"Successfully generated audio at: {output}")
        except Exception as e:
            print(f"Error using profile: {e}")

    elif args.command == "delete-profile":
        # Check manually first to avoid loading model if possible, 
        # or just use cloner. VoiceCloner init loads model which is slow.
        # But for 'delete', we don't need the model. 
        # Ideally we refactor VoiceCloner, but for now we can just use shutil directly here or instantiate cloner.
        # Instantiating is slow (loads model). Let's do a direct removal for CLI speed, 
        # OR fix VoiceCloner lazy loading.
        # Let's fix VoiceCloner lazy loading in a separate step if needed. 
        # For now, to suffice the user request without major refactor:
        if os.path.exists(os.path.join("profiles", args.name)):
            import shutil
            shutil.rmtree(os.path.join("profiles", args.name))
            print(f"Profile '{args.name}' deleted successfully.")
        else:
            print(f"Profile '{args.name}' not found.")

if __name__ == "__main__":
    main()