mateuseap/magic-vocals

1import ffmpeg2import numpy as np3 4# import praatio5# import praatio.praat_scripts6import os7import sys8 9import random10 11import csv12 13platform_stft_mapping = {14    "linux": "stftpitchshift",15    "darwin": "stftpitchshift",16    "win32": "stftpitchshift.exe",17}18 19stft = platform_stft_mapping.get(sys.platform)20# praatEXE = join('.',os.path.abspath(os.getcwd()) + r"\Praat.exe")21 22 23def CSVutil(file, rw, type, *args):24    if type == "formanting":25        if rw == "r":26            with open(file) as fileCSVread:27                csv_reader = list(csv.reader(fileCSVread))28                return (29                    (csv_reader[0][0], csv_reader[0][1], csv_reader[0][2])30                    if csv_reader is not None31                    else (lambda: exec('raise ValueError("No data")'))()32                )33        else:34            if args:35                doformnt = args[0]36            else:37                doformnt = False38            qfr = args[1] if len(args) > 1 else 1.039            tmb = args[2] if len(args) > 2 else 1.040            with open(file, rw, newline="") as fileCSVwrite:41                csv_writer = csv.writer(fileCSVwrite, delimiter=",")42                csv_writer.writerow([doformnt, qfr, tmb])43    elif type == "stop":44        stop = args[0] if args else False45        with open(file, rw, newline="") as fileCSVwrite:46            csv_writer = csv.writer(fileCSVwrite, delimiter=",")47            csv_writer.writerow([stop])48 49 50def load_audio(file, sr, DoFormant, Quefrency, Timbre):51    converted = False52    DoFormant, Quefrency, Timbre = CSVutil("csvdb/formanting.csv", "r", "formanting")53    try:54        # https://github.com/openai/whisper/blob/main/whisper/audio.py#L2655        # This launches a subprocess to decode audio while down-mixing and resampling as necessary.56        # Requires the ffmpeg CLI and `ffmpeg-python` package to be installed.57        file = (58            file.strip(" ").strip('"').strip("\n").strip('"').strip(" ")59        )  # 防止小白拷路径头尾带了空格和"和回车60        file_formanted = file.strip(" ").strip('"').strip("\n").strip('"').strip(" ")61 62        # print(f"dofor={bool(DoFormant)} timbr={Timbre} quef={Quefrency}\n")63 64        if (65            lambda DoFormant: True66            if DoFormant.lower() == "true"67            else (False if DoFormant.lower() == "false" else DoFormant)68        )(DoFormant):69            numerator = round(random.uniform(1, 4), 4)70            # os.system(f"stftpitchshift -i {file} -q {Quefrency} -t {Timbre} -o {file_formanted}")71            # print('stftpitchshift -i "%s" -p 1.0 --rms -w 128 -v 8 -q %s -t %s -o "%s"' % (file, Quefrency, Timbre, file_formanted))72 73            if not file.endswith(".wav"):74                if not os.path.isfile(f"{file_formanted}.wav"):75                    converted = True76                    # print(f"\nfile = {file}\n")77                    # print(f"\nfile_formanted = {file_formanted}\n")78                    converting = (79                        ffmpeg.input(file_formanted, threads=0)80                        .output(f"{file_formanted}.wav")81                        .run(82                            cmd=["ffmpeg", "-nostdin"],83                            capture_stdout=True,84                            capture_stderr=True,85                        )86                    )87                else:88                    pass89 90            file_formanted = (91                f"{file_formanted}.wav"92                if not file_formanted.endswith(".wav")93                else file_formanted94            )95 96            print(f" · Formanting {file_formanted}...\n")97 98            os.system(99                '%s -i "%s" -q "%s" -t "%s" -o "%sFORMANTED_%s.wav"'100                % (101                    stft,102                    file_formanted,103                    Quefrency,104                    Timbre,105                    file_formanted,106                    str(numerator),107                )108            )109 110            print(f" · Formanted {file_formanted}!\n")111 112            # filepraat = (os.path.abspath(os.getcwd()) + '\\' + file).replace('/','\\')113            # file_formantedpraat = ('"' + os.path.abspath(os.getcwd()) + '/' + 'formanted'.join(file_formanted) + '"').replace('/','\\')114            # print("%sFORMANTED_%s.wav" % (file_formanted, str(numerator)))115 116            out, _ = (117                ffmpeg.input(118                    "%sFORMANTED_%s.wav" % (file_formanted, str(numerator)), threads=0119                )120                .output("-", format="f32le", acodec="pcm_f32le", ac=1, ar=sr)121                .run(122                    cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True123                )124            )125 126            try:127                os.remove("%sFORMANTED_%s.wav" % (file_formanted, str(numerator)))128            except Exception:129                pass130                print("couldn't remove formanted type of file")131 132        else:133            out, _ = (134                ffmpeg.input(file, threads=0)135                .output("-", format="f32le", acodec="pcm_f32le", ac=1, ar=sr)136                .run(137                    cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True138                )139            )140    except Exception as e:141        raise RuntimeError(f"Failed to load audio: {e}")142 143    if converted:144        try:145            os.remove(file_formanted)146        except Exception:147            pass148            print("couldn't remove converted type of file")149        converted = False150 151    return np.frombuffer(out, np.float32).flatten()152