Function main

funasr/cli.py:81–167

()

def main():
    p = argparse.ArgumentParser(
        prog="funasr",
        description="FunASR - speech recognition CLI. 50+ languages, speaker diarization.",
        epilog="Examples:\n"
               "  funasr audio.wav\n"
               "  funasr audio.wav --model sensevoice -f json\n"
               "  funasr audio.wav -f srt -o ./subs\n"
               "  funasr audio.wav --spk --timestamps\n",
        formatter_class=argparse.RawDescriptionHelpFormatter,
    )
    p.add_argument("audio", nargs="+", help="Audio file(s) to transcribe")
    p.add_argument("--model", "-m", default="sensevoice", choices=list(MODEL_CONFIGS), help="Model (default: sensevoice)")
    p.add_argument("--language", "-l", default=None, help="Language: zh, en, ja, ko, yue, auto")
    p.add_argument("--device", default=None, help="Device: cuda:0, cpu (default: auto)")
    p.add_argument("--output-format", "-f", default="text", choices=["text", "json", "srt", "tsv"], help="Output format (default: text)")
    p.add_argument("--output-dir", "-o", default=None, help="Write output files to directory")
    p.add_argument("--timestamps", action="store_true", help="Include word-level timestamps")
    p.add_argument("--spk", action="store_true", help="Enable speaker diarization")
    p.add_argument("--hotwords", default=None, help="Comma-separated hotwords")
    p.add_argument("--verbose", "-v", action="store_true", help="Show loading/timing info on stderr")
    p.add_argument("--version", action="version", version=f"%(prog)s {_get_version()}")
    args = p.parse_args()

    if args.verbose:
        print(f"Loading model: {args.model} ...", file=sys.stderr)

    import torch
    from funasr import AutoModel

    device = args.device or ("cuda:0" if torch.cuda.is_available() else "cpu")
    config = MODEL_CONFIGS[args.model].copy()
    if args.spk and "spk_model" not in config:
        config["spk_model"] = "cam++"
    if "punc_model" not in config and args.model not in ("fun-asr-nano", "sensevoice"):
        config["punc_model"] = "ct-punc"

    t_load = time.time()
    model = AutoModel(device=device, disable_update=True, **config)
    if args.verbose:
        print(f"Model loaded in {time.time() - t_load:.1f}s", file=sys.stderr)

    if args.output_dir:
        os.makedirs(args.output_dir, exist_ok=True)

    for audio_path in args.audio:
        if not os.path.isfile(audio_path):
            print(f"Error: file not found: {audio_path}", file=sys.stderr)
            sys.exit(1)

        if args.verbose:
            print(f"Transcribing: {audio_path}", file=sys.stderr)

        t0 = time.time()
        gen_kw = {"input": audio_path, "batch_size": 1}
        if args.language:
            gen_kw["language"] = args.language
        if args.hotwords: