AI Music

AI Music & Audio Tools

Umbrella covering music generation, songwriting, and audio analysis. Each area has a dedicated reference file for detailed commands.

Contents

1. HeartMuLa — Music Generation from Lyrics + Tags
2. Songwriting & AI Music Prompts (Suno)
3. Audio Spectrogram & Feature Visualization

1. HeartMuLa — Music Generation (Lyrics + Tags)

Full reference: references/heartmula.md

HeartMuLa is an open-source (Apache-2.0) music foundation model family that generates songs from lyrics and style tags. Comparable to Suno for open-source.

Prerequisites

git clone https://huggingface.co/heartmula/HeartMuLa-3B
# Or use via HuggingFace Inference API

Quick Start

from heartmula import HeartMuLa
model = HeartMuLa.from_pretrained("heartmula/HeartMuLa-3B")
audio = model.generate(lyrics="Verse 1...", tags="pop, upbeat, female vocals")

Supported Tags

• Genre: pop, rock, jazz, classical, electronic, hip-hop, r&b, folk, metal, blues, country, latin, ambient, lo-fi
• Mood: happy, sad, energetic, calm, dark, uplifting, dreamy, aggressive, romantic, nostalgic
• Instrument: piano, guitar, drums, strings, synth, saxophone, flute, organ, bass, orchestra
• Vocal: male, female, duet, choir, rap, spoken, falsetto, harmonies, acapella
• Tempo: slow, medium, fast, bpm:120

2. Songwriting & AI Music Prompts

Full reference: references/songwriting.md

Crafting Lyrics

Golden rules:

• Show, don't tell — use concrete imagery instead of abstract emotion
• Syllable count matters more than rhyme
• A strong hook repeats 3-4 times
• Write like you talk, then polish

Song Structure

Intro (4-8 bars) → Verse (16 bars) → Chorus (8 bars) → Verse → Chorus → Bridge (8 bars) → Chorus (×2) → Outro

Suno Prompt Patterns

Parody & Adaptation

• Match syllable count and stress pattern of the original
• Keep the original rhyme scheme
• Use phonetic tricks for rhymes

3. Audio Spectrogram & Feature Visualization

Full reference: references/songsee.md

Use songsee CLI to generate spectrograms and audio feature visualizations from audio files.

Quick Start

# Install
pip install songsee

# Generate mel spectrogram
songsee spectrogram input.mp3 -o output.png

# Features
songsee features input.mp3 --chroma --mfcc --tempogram

Supported Visualizations

• Mel spectrogram — frequency content over time
• Chroma — pitch class distribution
• MFCC — timbral features (speech/music recognition)
• Tempogram — tempo and rhythmic patterns
• Spectral features — centroid, bandwidth, rolloff, flatness

Use Cases

• Analyzing song structure (verse/chorus boundaries from spectral changes)
• Debugging audio processing pipelines
• Visual documentation for music production
• Comparing original vs generated audio quality