vault backup: 2023-12-12 18:22:47

.obsidian/workspace.json

@@ -224,9 +224,13 @@
   },
   "active": "540ef396474d1c8b",
   "lastOpenFiles": [
-    "05. 資料收集/libsndfile.md",
+    "00. Inbox/使用 librosa 做 mel spectrogram.md",
+    "attachments/Pasted image 20231212181946.png",
     "00. Inbox/使用 librosa 做 FFT.md",
-    "00. Inbox/mel spectrogram.md",
+    "attachments/Pasted image 20231212181543.png",
+    "00. Inbox/Mel spectrogram.md",
+    "attachments/Pasted image 20231212180917.png",
+    "05. 資料收集/libsndfile.md",
     "00. Inbox/SSIM.md",
     "00. Inbox/Brown noise.md",
     "00. Inbox/White noise.md",
@@ -251,8 +255,6 @@
     "00. Inbox/01. TODO.md",
     "05. 資料收集/架站/Storj.md",
     "01. 個人/01. Daily/2023/11/2023-11-23(週四).md",
-    "00. Inbox/Habit Tracker.md",
-    "01. 個人/01. Daily/2023/11/2023-11-22(週三).md",
     "01. 個人/01. Daily/2023/10",
     "00. Inbox/vc-fwUpdate code trace(Meetup).canvas",
     "01. 個人/01. Daily/2023/09",
@@ -270,9 +272,6 @@
     "attachments/Pasted image 20230426213640.png",
     "attachments/Pasted image 20230426213616.png",
     "attachments/Pasted image 20230426213611.png",
-    "attachments/Pasted image 20230426213547.png",
-    "attachments/Pasted image 20230426213534.png",
-    "attachments/Pasted image 20230426213513.png",
-    "attachments/Pasted image 20230418111538.png"
+    "attachments/Pasted image 20230426213547.png"
   ]
 }

00. Inbox/使用 librosa 做 FFT.md

@@ -0,0 +1,27 @@
+
+```python
+import numpy as np
+n_fft=2048
+ft = np.abs(librosa.stft(y[:n_fft], hop_length=n_fft+l))
+plt.plot(ft)
+plt.title('Spectrum')
+plt.xlabel('Frequency Bin')
+plt.ylabel('Amp1itude')
+```
+
+output e.g.:
+![[Pasted image 20231212180917.png]]
+
+這時候 Y 軸是 amplitude，可以使用 `librosa.amplitude_to_db()` 來把 amplitude 轉為 db。
+
+```python
+import librosa.display
+spec = np.abs(librosa.stft(y, hop_1ength=512))
+spec = librosa.amplitude_to_db(spec, ref=np.max)
+librosa.display.specshow(spec, sr=sr, x_axis='time', y_axis='log')
+p1t.colorbar(format='%+2.0f dB')
+plt.title( 'Spectrogram')
+```
+
+Output:
+![[Pasted image 20231212181543.png]]

00. Inbox/使用 librosa 做 mel spectrogram.md

@@ -0,0 +1,13 @@
+由於人類會對低頻低音高的片段更感興趣，所以會對通過 FFT 變換得到的 Amplitude 和 Frequency。
+[[Mel spectrogram]] 和 spectrogram 的差別就是 mel spectrogram 的頻率是 mel scale 變換後的頻率（你可以想像把Spectrogram整體往下壓）
+
+```python
+mel_spect = librosa.feature.melspectrogram(y=y, sr=sr, n fft=2048, hop_Iength=1024)
+mel_spect = librosa.power_to_db(mel_spect, ref=np.max)
+librosa.display.specshow(mel_spect, y_axis='mel', fmax=8000, x_axis='time')
+plt.title('Mel Spectrogram')
+p1t.colorbar(format='%+2.0f dB')
+```
+
+Output:
+![[Pasted image 20231212181946.png]]