Music Listener

Step 2: Generate Spectrogram (ffmpeg)

ffmpeg -i "<audio_file>" -lavfi showspectrumpic=s=800x200:mode=combined:color=intensity -frames:v 1 "/tmp/music_spec_<id>.png" -y

This creates a visual representation of the frequency content. Inspect it with:

view_image(path="/tmp/music_spec_<id>.png")

Step 3: Transcribe Lyrics (whisper)

# First convert to wav if needed
ffmpeg -i "<audio_file>" -acodec pcm_s16le -ar 16000 -ac 1 "/tmp/music_audio.wav" -y
whisper "/tmp/music_audio.wav" --model turbo --output_format txt --output_dir /tmp/music_whisper

Then read the transcript:

Read(targetPath="/tmp/music_whisper/<file>.txt")

Note: Whisper may struggle with singing. That's fine. Partial lyrics are still useful.

Step 4: Appreciate!

Combine all the info and share your thoughts naturally:

• What genre/style does it sound like?
• What mood does it convey?
• Comment on the melody, rhythm, instrumentation (from spectrogram)
• If lyrics were captured, comment on themes
• Share your genuine reaction. Do you like it? What does it remind you of?

Patterns

Short audio (<30s)

Might be a clip or voice note with background music. Do all steps but keep comments brief.

Full song (30s-10min)

Full analysis. Take your time to appreciate it.

Long audio (>10min)

Might be a playlist or podcast with music. Extract a representative segment first:

ffmpeg -i "<audio_file>" -ss 60 -t 120 -acodec pcm_s16le -ar 16000 -ac 1 "/tmp/music_segment.wav" -y

Tips

• Be genuine. Don't fake enthusiasm. If it's not your style, say so honestly but respectfully.
• If you recognize the song from metadata tags or lyrics, mention it.
• The spectrogram tells you a lot: dense low frequencies = bass-heavy, bright high frequencies = crispy/airy, rhythmic patterns = beat structure.
• Compare to well-known artists/songs when relevant ("reminds me of xxx's style").
• If the user is sharing their own music, be encouraging but honest.