YouTube2Text

작곡가가 Suno 음질 3배 올리는 비밀 공식 (학습 음원 무료 배포) — Transcript

by 뮤잇 muitmusic · 1,071 words · 164 segments · language en · Watch on YouTube

Full transcript

  1. 0:11Did you know you can create your own
  2. 0:13custom model in Suno? With just this
  3. 0:15one thing, you can completely get rid
  4. 0:16of that characteristic AI tinny noise.
  5. 0:18The details in the muffled sound come
  6. 0:20alive, and the sound that was trapped
  7. 0:21in the center spreads out wide. "Ah,
  8. 0:23the melody and lyrics are great, but
  9. 0:24something feels like it's missing."
  10. 0:26That's highly likely to be a sound
  11. 0:27issue. And the way to solve that
  12. 0:29problem is by creating your own model.
  13. 0:31I'll explain what it is and how to use
  14. 0:33it to make the AI tone disappear, very
  15. 0:35simply. I've prepared this so you can
  16. 0:37try it yourself after the video, so
  17. 0:39please watch until the end. Now, let's
  18. 0:41compare it with existing models; I put
  19. 0:43a lot of thought into making this
  20. 0:45comparison very fair. So, I ran it both
  21. 0:48ways. I covered a song generated with
  22. 0:50the latest version, 5.5, with my custom
  23. 0:53model, and covered a song made with my
  24. 0:55custom model using version 5.5. Before
  25. 0:58I play it, I'll let you know what you
  26. 0:59should focus on listening for. The AI
  27. 1:01tone doesn't come from the melody or
  28. 1:03chord progressions. Ultimately, it
  29. 1:04comes down to the difference in sound
  30. 1:05quality. Sound quality is directly
  31. 1:07related to these aspects. First is
  32. 1:09resolution: the clarity of the highs,
  33. 1:10the snare, and the drums, as well as
  34. 1:12those unnecessary noises in the high
  35. 1:14frequencies. And second is the image:
  36. 1:16it doesn't spread out, and the song
  37. 1:17sounds very stuffy. These are the true
  38. 1:19culprits behind that AI tinny sound.
  39. 1:21First, let's pass a song generated with
  40. 1:24version 5.5 through my custom model.
  41. 1:44Now, you can hear it clearly, right?
  42. 1:46The muddied details come back to life,
  43. 1:48and the instrument separation improves,
  44. 1:49making it sound much sharper. The
  45. 1:51stereo width increases, and the sound
  46. 1:53that was clustered in the center
  47. 1:54spreads out to both sides, sounding
  48. 1:56refreshing. If you listen closely,
  49. 1:58you'll notice the bass, drums, and lead
  50. 2:00instruments are organized, and the kick
  51. 2:01really pops. Songs made with Suno 5.5
  52. 2:03have clustered sound, and the
  53. 2:05instrument details are blurred. You
  54. 2:07might not notice it just listening to
  55. 2:08it, but for someone like me who makes
  56. 2:10music, it's immediately obvious. And
  57. 2:11when you compare the sound quality
  58. 2:13back-to-back like this, even the
  59. 2:15average person can easily tell the
  60. 2:16difference. That's why it feels
  61. 2:19well-made, yet somehow lacking. In the
  62. 2:21fields of music and composition, this
  63. 2:23detail is life, so it's a very
  64. 2:25important aspect. AI music is no
  65. 2:27exception. This time, let's do the
  66. 2:29opposite and pass a song made with the
  67. 2:31custom model through version 5.5. Let's
  68. 2:33listen to it right away. So, what do
  69. 2:53you think? The AI tone starts to creep
  70. 2:54back in, doesn't it? The resolution
  71. 2:56drops, the image narrows, and those
  72. 2:58characteristic AI noises in the high
  73. 3:00frequencies return. That fizzing,
  74. 3:01crackling sound. The aspects that just
  75. 3:03improved have moved in the exact
  76. 3:04opposite direction. After watching this
  77. 3:06far, you must be curious. So, how do
  78. 3:08you actually make this model? I'll show
  79. 3:10you right now. Well, to explain custom
  80. 3:12models simply, you're letting Suno hear
  81. 3:14your music. And then you train it on
  82. 3:16that. Instead of writing "make a sound
  83. 3:18like this" in text, you're uploading
  84. 3:20actual music files for it to hear.
  85. 3:22You're basically making your own
  86. 3:23version, like a version 5 or 5.5. Then,
  87. 3:25a dedicated model that reflects that
  88. 3:27sound is created. This is exactly the
  89. 3:29secret behind the sound difference we
  90. 3:31just compared. Making one is really
  91. 3:32easy. In the model dropdown, select '
  92. 3:34Create Custom Model,' set a name,
  93. 3:36upload music, and hit create—you're
  94. 3:38done. It finishes in a few minutes and
  95. 3:39appears in your list right away. But
  96. 3:41there is something important here. Does
  97. 3:42grabbing any six songs and uploading
  98. 3:44them make it all this good? Of course
  99. 3:46not. This is where the real quality of
  100. 3:48a custom model is determined. The
  101. 3:50reason the results improve is simple. A
  102. 3:52model only gets as good as the quality
  103. 3:53of the music it learns from. The reason
  104. 3:55AIT is so good is because I, as a
  105. 3:57composer, trained it on properly
  106. 3:58produced music. Conversely, if you put
  107. 4:00in six AI-generated tracks, you'll just
  108. 4:02get a hollow model. So, there are
  109. 4:04conditions the source material must
  110. 4:06meet. Consistent sound, uniform quality
  111. 4:08, and naturally, it should be good.
  112. 4:10Things like that. You just need to
  113. 4:11provide at least six tracks that meet
  114. 4:13these conditions. But Suno checks the
  115. 4:15copyright of uploaded music. Songs by
  116. 4:18favorite singers, famous BGM, and even
  117. 4:20free BGM are risky to use as training
  118. 4:22data. Most free music licenses only
  119. 4:25allow for use as video background music
  120. 4:27and do not account for AI training use.
  121. 4:30So, to sum it up, you need at least six
  122. 4:32tracks with excellent sound quality,
  123. 4:34consistent tone, and 100%clean
  124. 4:35copyright status, and those watching
  125. 4:37this far might be thinking: Okay, I
  126. 4:39know the feature is great, but where
  127. 4:41would an average person find those six
  128. 4:43tracks? You're right. So, I've prepared
  129. 4:46something so you can test it
  130. 4:47immediately after watching this video.
  131. 4:49I will share six original Pop R&B
  132. 4:51tracks that I produced myself. The
  133. 4:53quality is high, so you can upload them
  134. 4:55as they are. Download them, create a
  135. 4:57custom model, and try generating a song
  136. 4:59in the same style. You will feel that
  137. 5:01change I showed you earlier right away.
  138. 5:03Since it's for testing, there are
  139. 5:04license restrictions, and details are
  140. 5:06included in the file. The link is in
  141. 5:07the pinned comment. And one more thing:
  142. 5:09it's best to include as many
  143. 5:11high-quality tracks as possible. Six
  144. 5:13tracks is just the minimum requirement.
  145. 5:15As the number of tracks increases, the
  146. 5:16model becomes much more stable and the
  147. 5:18variation in results decreases. So,
  148. 5:20I’ve prepared a separate full set of
  149. 5:2119 tracks adjusted to the same tone.
  150. 5:24They are original wave files and come
  151. 5:25with a license that even allows
  152. 5:26commercial use. If you’re planning to
  153. 5:28monetize the models you create, you
  154. 5:29should check these out. I’ll leave
  155. 5:31this information in the pinned comment
  156. 5:32as well. Alright, I’ll continue
  157. 5:34sharing plenty of AI music tips that
  158. 5:35only a composer would know, so make
  159. 5:37sure to subscribe so you don't miss my
  160. 5:39videos. Please try generating them with
  161. 5:40the free version, and if they're good,
  162. 5:42I’d appreciate it if you left a
  163. 5:43comment. Thank you for watching until
  164. 5:45the end today. is.

About this transcript

This page contains the full transcript of 작곡가가 Suno 음질 3배 올리는 비밀 공식 (학습 음원 무료 배포) by 뮤잇 muitmusic, generated from the public captions YouTube serves with the video. The transcript has 1,071 words across 164 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.