devbox
Li_阴宅
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

RNN生成音乐_自动生成音乐命令示例

作者：Li_阴宅 | 2024-07-13 10:41:33

赞

踩

自动生成音乐命令示例

计算图模型


1.    下载midi文件

 下载midi格式音乐的网站：freemidi.org


2.创建NoteSequences

即创建旋律数据库，将MIDI集合转化为NoteSequences。（NoteSequences是协议缓冲区，它是一种快速有效的数据格式，并且比MIDI文件更易于使用）

  1. INPUT_DIRECTORY=/Users/mac/Desktop/MIDI
  2.  
  3. SEQUENCES_TFRECORD=/Users/mac/Desktop/train/notesequences.tfrecord
  4.  
  5.  
  6.  
  7. convert_dir_to_note_sequences\
  8.  
  9.   --input_dir=$INPUT_DIRECTORY \
  10.  
  11.   --output_file=$SEQUENCES_TFRECORD \
  12.  
  13.   --recursive

注意：路径中不能有空格

显示如下图则操作成功：


  1. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$INPUT_DIRECTORY=/Users/mac/Desktop/MIDI
  2.  
  3. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$ # TFRecord file that will containNoteSequence protocol buffers.
  4.  
  5. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$SEQUENCES_TFRECORD=/Users/mac/Desktop/train/notesequences.tfrecord
  6.  
  7. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$
  8.  
  9. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$ convert_dir_to_note_sequences \
  10.  
  11. >   --input_dir=$INPUT_DIRECTORY \
  12.  
  13. >   --output_file=$SEQUENCES_TFRECORD \
  14.  
  15. >   --recursive
  16.  
  17. INFO:tensorflow:Convertingfiles in '/Users/mac/Desktop/MIDI/'.
  18.  
  19. /Users/mac/miniconda2/envs/magenta/lib/python2.7/site-packages/pretty_midi/pretty_midi.py:100:RuntimeWarning: Tempo, Key or Time signature change events found on non-zerotracks. This is not a valid type 0 ortype 1 MIDI file.  Tempo, Key or TimeSignature may be wrong.
  20.  
  21.   RuntimeWarning)
  22.  
  23. INFO:tensorflow:Converted70 files in '/Users/mac/Desktop/MIDI/'.
  24.  
  25. INFO:tensorflow:Could not parse 0 files.
  26.  
  27. INFO:tensorflow:Wrote70 NoteSequence protos to '/Users/mac/Desktop/train/notesequences.tfrecord'
  28.  
  29. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$

3.创建SequenceExamples

SequenceExamples被提供给模型用来训练和评估。每个SequenceExample包含一个输入序列和代表一个旋律的一系列标签。下面的命令是从NoteSequences中提取旋律，并将它们保存为SequenceExamples。将生成两个SequenceExamples集合，一个用于训练，另一个用于评估，其中评估集中SequenceExamples的分数由-eval_ratio确定。评估率为0.10，提取的旋律的10％将保存在评估系统中，90％将被保存在训练集中。

 

代码：

  1. melody_rnn_create_dataset\
  2.  
  3. --config=lookback_rnn\
  4.  
  5. --input=/Users/mac/Desktop/train/notesequences.tfrecord\
  6.  
  7. --output_dir=/Users/mac/Desktop/train/melody_rnn/sequence_examples\
  8.  
  9. --eval_ratio=0.10

代码运行后生成的文件


显示如下图则操作成功：

  1. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$ melody_rnn_create_dataset \
  2.  
  3. >--config=lookback_rnn \
  4.  
  5. >--input=/Users/mac/Desktop/train/notesequences.tfrecord \
  6.  
  7. >--output_dir=/Users/mac/Desktop/train/melody_rnn/sequence_examples \
  8.  
  9. >--eval_ratio=0.10
  10.  
  11. INFO:tensorflow:
  12.  
  13.  
  14.  
  15. Completed.
  16.  
  17.  
  18.  
  19. INFO:tensorflow:Processed70 inputs total. Produced 264 outputs.
  20.  
  21. INFO:tensorflow:DAGPipeline_MelodyExtractor_eval_melodies_discarded_too_few_pitches:4
  22.  
  23. INFO:tensorflow:DAGPipeline_MelodyExtractor_eval_melodies_discarded_too_long:0
  24.  
  25. INFO:tensorflow:DAGPipeline_MelodyExtractor_eval_melodies_discarded_too_short:40
  26.  
  27. INFO:tensorflow:DAGPipeline_MelodyExtractor_eval_melodies_truncated:2
  28.  
  29. INFO:tensorflow:DAGPipeline_MelodyExtractor_eval_melody_lengths_in_bars:
  30.  
  31.   [7,8): 1
  32.  
  33.   [8,10): 1
  34.  
  35.   [10,20): 6
  36.  
  37.   [20,30): 3
  38.  
  39.   [30,40): 1
  40.  
  41. INFO:tensorflow:DAGPipeline_MelodyExtractor_eval_polyphonic_tracks_discarded:34
  42.  
  43. INFO:tensorflow:DAGPipeline_MelodyExtractor_training_melodies_discarded_too_few_pitches:87
  44.  
  45. INFO:tensorflow:DAGPipeline_MelodyExtractor_training_melodies_discarded_too_long:0
  46.  
  47. INFO:tensorflow:DAGPipeline_MelodyExtractor_training_melodies_discarded_too_short:1001
  48.  
  49. INFO:tensorflow:DAGPipeline_MelodyExtractor_training_melodies_truncated:54
  50.  
  51. INFO:tensorflow:DAGPipeline_MelodyExtractor_training_melody_lengths_in_bars:
  52.  
  53.   [7,8): 38
  54.  
  55.   [8,10): 28
  56.  
  57.   [10,20): 92
  58.  
  59.   [20,30): 39
  60.  
  61.   [30,40): 52
  62.  
  63.   [40,50): 3
  64.  
  65. INFO:tensorflow:DAGPipeline_MelodyExtractor_training_polyphonic_tracks_discarded:744
  66.  
  67. INFO:tensorflow:DAGPipeline_RandomPartition_eval_melodies_count:4
  68.  
  69. INFO:tensorflow:DAGPipeline_RandomPartition_training_melodies_count:66
  70.  
  71. (magenta)zhu-Macs-MacBook-Pro:magenta zhujianing$

4.训练并测试模型

  1. melody_rnn_train\
  2.  
  3. --config=lookback_rnn\
  4.  
  5. --run_dir=/Users/mac/Desktop/train/melody_rnn/logdir/run1\
  6.  
  7. --sequence_example_file=/Users/mac/Desktop/train/melody_rnn/sequence_examples/training_melodies.tfrecord\
  8.  
  9. --hparams="{'batch_size':64,'rnn_layer_sizes':[64,64]}"\
  10.  
  11. --num_training_steps=20000

5. 根据开头给定的旋律给出midi格式音乐

  1. melody_rnn_generate\
  2.  
  3. --config=lookback_rnn\
  4.  
  5. --run_dir=/Users/mac/Desktop/train/melody_rnn/logdir/run1\
  6.  
  7. --output_dir=/Users/mac/Desktop/train/melody_rnn/generated\
  8.  
  9. --num_outputs=10\
  10.  
  11. --num_steps=512\
  12.  
  13. --hparams="{'batch_size':64,'rnn_layer_sizes':[64,64]}"\
  14.  
  15. --primer_midi=/Users/mac/Desktop/MIDI/21Guns.mid

播放midi文件方式

  1. $ brew install timidity  
  2.  
  3. $timidity /Users/mac/Desktop/MIDI/21Guns\(1\).mid

结果如下图就可以成功播放mid文件了。



声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/Li_阴宅/article/detail/819064
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号