使用python进行节奏检测

Question

我想用python检测给定音乐或声音的节奏。 然后将节奏时间点打印到文件中，以便在我的统一游戏中使用。 我已经在这方面取得了一些进展，但它仍然不是我想要的。 它缺少一些注释。 这是我的python代码。 我目前正在为此使用librosa模块。 我的问题是，我走对了路吗？ 如果是，那么我该如何改进我的代码？ 如果没有，那我该怎么办？

import librosa
import numpy as np
from os import listdir, path, remove
from os.path import isfile, join
from pydub import AudioSegment

SONGS_PATH = "./songs"
PS_FILES_PATH = "./ps_files"
detection_mode = "full"

# Get the song files from given dir
song_files = [f for f in listdir(SONGS_PATH) if isfile(join(SONGS_PATH, f))]


count = len(song_files)
# Process each song file
for i, song_file in enumerate(song_files):
    print(str(i+1) + "/" + str(count) + " Processing song: ", song_file)
    song_file_path = SONGS_PATH + "/" + song_file
    output_file = PS_FILES_PATH + "/" + song_file.replace(".mp3", "").replace(".wav", "")

    # If file type is mp3 then convert it to wav and remove mp3 file
    if ".mp3" in song_file:
        print("Not supported format. Converting to wav...")
        sound = AudioSegment.from_mp3(song_file_path)
        sound.export(song_file_path.replace("mp3", "wav"), format="wav")
        print("File converted. Deleting old one...")
        remove(song_file_path)
        song_file_path = song_file_path.replace("mp3", "wav")

    # Load file
    xy, sr = librosa.load(song_file_path)
    # Find onsets
    onset_frames = librosa.onset.onset_detect(xy, sr=sr, wait=1, pre_avg=1, post_avg=1, pre_max=1, post_max=1)
    # Get the onset times
    onset_times = librosa.frames_to_time(onset_frames)
    onset_clicks = librosa.clicks(frames=onset_frames, sr=sr, length=len(xy))

    # find tempo and beats
    tempo, beat_frames = librosa.beat.beat_track(y=xy, sr=sr)
    # Get the beat times
    beat_times = librosa.frames_to_time(beat_frames)
    beat_clicks = librosa.clicks(frames=beat_frames, sr=sr, length=len(xy))

    # Generate a txt file which is contains times, and create a audio file with click effect.
    print("Generating ps file...")
    if detection_mode == "only-onsets":
        librosa.output.write_wav(output_file + "onset_clicks.wav", xy + onset_clicks, sr)
        file = open(output_file + "_onset_times.txt","ab")
        np.savetxt(file, onset_times, '%.2f') 
    elif detection_mode == "only-beats":
        librosa.output.write_wav(output_file + "beat_clicks.wav", xy + beat_clicks, sr)
        file = open(output_file + "_beat_times.txt","ab")
        np.savetxt(file, beat_times, '%.2f') 
    else:
        librosa.output.write_wav(output_file + "full_clicks.wav", xy + onset_clicks + beat_clicks, sr)
        file = open(output_file + "_full_times.txt","ab")
        np.savetxt(file, np.concatenate(beat_times, onset_times), '%.2f') 
    print("Done.\n")

这是我生成的具有点击效果的声音文件之一： 声音文件

编辑：

我实际上是在做一个游戏，比如钢琴砖，吉他英雄。 但是我的游戏不会包含 3 行，它只有一行用于击中音符，这意味着我只需要一个事件来触发我的机制。

这是我想要的事件机制的示例视频： 示例视频