claude-music-analyze

claude-music-analyze — Audio Analysis

Quick Analysis (ffprobe)

# Full metadata
ffprobe -v quiet -print_format json -show_format -show_streams input.flac

# Duration only
ffprobe -v quiet -show_entries format=duration -of csv=p=0 input.flac

# Sample rate, channels, codec
ffprobe -v quiet -show_entries stream=sample_rate,channels,codec_name -of csv=p=0 input.flac

Loudness Measurement (LUFS)

ffmpeg -i input.flac -af loudnorm=I=-14:TP=-1:LRA=11:print_format=json -f null - 2>&1 | \
  grep -A 20 '"input_'

Key fields in output:

• input_i: Integrated loudness (LUFS)
• input_tp: True peak (dBTP)
• input_lra: Loudness range (LU)

BPM Detection

Using ffmpeg's ebur128 for rhythm analysis, or for accurate BPM:

# If librosa is available in ~/.video-skill/ venv:
source ~/.video-skill/bin/activate
python3 -c "
import librosa
y, sr = librosa.load('input.flac', sr=None)
tempo, _ = librosa.beat.beat_track(y=y, sr=sr)
print(f'BPM: {tempo[0]:.1f}' if hasattr(tempo, '__len__') else f'BPM: {tempo:.1f}')
"

Key Detection

source ~/.video-skill/bin/activate
python3 -c "
import librosa
import numpy as np
y, sr = librosa.load('input.flac', sr=None)
chroma = librosa.feature.chroma_cqt(y=y, sr=sr)
key_names = ['C', 'C#', 'D', 'D#', 'E', 'F', 'F#', 'G', 'G#', 'A', 'A#', 'B']
key_idx = np.argmax(np.mean(chroma, axis=1))
print(f'Estimated key: {key_names[key_idx]}')
"

Comprehensive Report

Combine all analyses into a single report:

1. Run ffprobe for format/codec/duration
2. Run loudnorm for LUFS measurement
3. Run librosa for BPM + key (if available)
4. Present as structured summary