How to reverse spectrogram png into audio

[John Ho]

def audio_to_spectrogram(audio,img):
  plt.rcParams["figure.figsize"] = [2.56, 2.56]
  plt.rcParams["figure.autolayout"] = True

  fig, ax = plt.subplots()


  y, sr = librosa.load(audio)
  S = librosa.feature.melspectrogram(y=y, sr=sr)
  print(sr)
  S_dB = librosa.power_to_db(S, ref=np.max)
  p = librosa.display.specshow(S_dB, sr=sr, fmax=8000, ax=ax)

  plt.savefig(img)

Using this code, I have converted my wav file into a 256x256 image to be used in a neural network. I was wondering how I could reverse the image back into audio.

[Graham Coleman]

Hi John,

As long as your image matches the dimensions of your mel spectrogram (and all of the parameters match), you can use the following function to invert it (using Griffin-Lim) back to audio:

https://librosa.org/doc/main/generated/librosa.feature.inverse.mel_to_audio.html

Graham

--
You received this message because you are subscribed to the Google Groups "librosa" group.
To unsubscribe from this group and stop receiving emails from it, send an email to librosa+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/librosa/3b84447b-6062-4506-b28d-34b89d4095fbn%40googlegroups.com.

[John Ho]

Hi, 

These are some example sizes of my mel-spectrogram

(128, 59)

(128, 60)

(128, 27)

(128, 438)

(128, 652)

However, when I plot it, the image becomes size (256,256,3).

How would i convert this array of (256,256,3) back into the dimensions of the spectrogram?