Songsee

Audio Visualization with Spectrograms

Visualize audio files as mel spectrograms, chromagrams, and MFCC feature plots using the librosa Python library. Useful for music analysis, speech processing, and audio debugging.

When to Use

• User wants to visualize what an audio file "looks like"
• User wants to analyze the frequency content of a recording
• User wants to compare two audio files visually
• User wants to understand musical key or chroma content
• User wants to extract MFCC features for a machine learning task

How to Use

1. Install dependencies

pip install librosa matplotlib soundfile

2. Generate a mel spectrogram

import librosa
import librosa.display
import matplotlib.pyplot as plt
import numpy as np

def mel_spectrogram(audio_path, output="mel_spec.png"):
  y, sr = librosa.load(audio_path, sr=None)
  S     = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=128, fmax=8000)
  S_db  = librosa.power_to_db(S, ref=np.max)

  fig, ax = plt.subplots(figsize=(12, 4), facecolor="#0d1117")
  ax.set_facecolor("#0d1117")
  img = librosa.display.specshow(S_db, sr=sr, x_axis="time", y_axis="mel", fmax=8000, ax=ax, cmap="magma")
  fig.colorbar(img, ax=ax, format="%+2.0f dB", label="dB")
  ax.set_title(f"Mel Spectrogram — {audio_path}", color="white")
  ax.tick_params(colors="white")
  ax.xaxis.label.set_color("white")
  ax.yaxis.label.set_color("white")
  plt.tight_layout()
  plt.savefig(output, dpi=150, bbox_inches="tight")
  plt.close()
  print(f"Saved: {output}")

mel_spectrogram("song.mp3")

3. Generate a chromagram (musical key content)

import librosa, librosa.display, matplotlib.pyplot as plt

def chromagram(audio_path, output="chroma.png"):
  y, sr   = librosa.load(audio_path, sr=None)
  chroma  = librosa.feature.chroma_cqt(y=y, sr=sr)

  fig, ax = plt.subplots(figsize=(12, 4), facecolor="#0d1117")
  ax.set_facecolor("#0d1117")
  img = librosa.display.specshow(chroma, y_axis="chroma", x_axis="time", ax=ax, cmap="coolwarm")
  fig.colorbar(img, ax=ax)
  ax.set_title("Chromagram", color="white")
  ax.tick_params(colors="white")
  plt.tight_layout()
  plt.savefig(output, dpi=150, bbox_inches="tight")
  plt.close()
  print(f"Saved: {output}")

chromagram("song.mp3")

4. Generate MFCC features

import librosa, librosa.display, matplotlib.pyplot as plt
import numpy as np

def mfcc_plot(audio_path, n_mfcc=20, output="mfcc.png"):
  y, sr = librosa.load(audio_path, sr=None)
  mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=n_mfcc)

  fig, ax = plt.subplots(figsize=(12, 4), facecolor="#0d1117")
  ax.set_facecolor("#0d1117")
  img = librosa.display.specshow(mfccs, x_axis="time", ax=ax, cmap="viridis")
  fig.colorbar(img, ax=ax)
  ax.set_title(f"MFCC ({n_mfcc} coefficients)", color="white")
  ax.tick_params(colors="white")
  plt.tight_layout()
  plt.savefig(output, dpi=150, bbox_inches="tight")
  plt.close()
  print(f"Saved: {output}")

mfcc_plot("speech.wav", n_mfcc=13)

5. Generate all three plots at once

def analyze_audio(audio_path):
  base = audio_path.rsplit(".", 1)[0]
  mel_spectrogram(audio_path, output=f"{base}_mel.png")
  chromagram(audio_path,      output=f"{base}_chroma.png")
  mfcc_plot(audio_path,       output=f"{base}_mfcc.png")
  print(f"Analysis complete: 3 PNG files saved for {audio_path}")

analyze_audio("recording.wav")

6. Get basic audio statistics

import librosa, numpy as np

y, sr     = librosa.load("audio.mp3", sr=None)
duration  = librosa.get_duration(y=y, sr=sr)
tempo, _  = librosa.beat.beat_track(y=y, sr=sr)
rms       = np.sqrt(np.mean(y**2))

print(f"Duration:   {duration:.2f} seconds")
print(f"Sample rate:{sr} Hz")
print(f"Tempo:      {tempo:.1f} BPM")
print(f"RMS energy: {rms:.4f}")

Examples

"Show me what this audio recording looks like as a spectrogram" → Use step 2 to generate a mel spectrogram PNG. Open the saved file.

"What musical key is this song in? Visualize the chroma content" → Use step 3 to generate a chromagram — peaks in chroma rows indicate dominant pitch classes.

"Generate MFCC features from this speech recording for my ML model" → Use step 4 to plot MFCCs, then extract the mfccs array for downstream ML use.

Cautions

• librosa loads audio in float32 mono by default — stereo files are mixed down automatically
• Large audio files (> 30 minutes) take significant time and memory to process — slice with offset and duration parameters if needed
• librosa.load supports MP3, WAV, FLAC, OGG — ensure soundfile and audioread are installed for MP3 support
• MFCC coefficients are sensitive to n_mfcc and sr — use consistent settings across all files in an ML dataset