app_cli.py

import argparse
import shutil
import soundfile as sf

import app
from app import infer


def main():
    parser = argparse.ArgumentParser(description="F5 TTS - CLI interface")
    parser.add_argument("--ref_audio", required=True, help="Path to the reference audio file (wav, mp3)")
    parser.add_argument("--ref_text", default="", help="Reference text. If empty, audio transcription will be performed.")
    parser.add_argument("--gen_text", required=True, help="Text to generate")
    parser.add_argument("--exp_name", default="Multi", help="Experiment name / model selection (default: Multi)")
    parser.add_argument("--remove_silence", action="store_true", help="Remove silence from the output audio")
    parser.add_argument("--cross_fade_duration", type=float, default=0.15, help="Cross-fade duration (s)")
    parser.add_argument("--output_audio", default="output.wav", help="Path to the output WAV file")
    parser.add_argument("--output_spectrogram", default="spectrogram.png", help="Path to save the spectrogram (PNG)")
    parser.add_argument("--language", default="en-us", help="Synthesized language (default: en-us)")
    parser.add_argument("--ref_language", default="en-us", help="Reference language (default: en-us)")
    parser.add_argument("--speed", type=float, default=1.0, help="Audio speed factor (default: 1.0)")
    args = parser.parse_args()

    app.language = args.language
    app.ref_language = args.ref_language
    app.speed = args.speed

    (sr, audio_data), spectrogram_path = infer(
        args.ref_audio,
        args.ref_text,
        args.gen_text,
        args.exp_name,
        args.remove_silence,
        args.cross_fade_duration
    )

    sf.write(args.output_audio, audio_data.astype("float32"), sr)
    shutil.copy(spectrogram_path, args.output_spectrogram)

    print(f"Audio saved in: {args.output_audio}")
    print(f"Spectrogram saved in: {args.output_spectrogram}")

if __name__ == "__main__":
    main()