SonicAI

Browser-Based AI Text-to-Song Generator

Transform text and lyrics into music — entirely in your browser, no server required.

---

🎯 Overview

SonicAI is a real-time text-to-song generation system that runs entirely client-side in the browser. It converts lyrics into musical compositions using the Web Audio API for instrument synthesis and formant vocal modeling for singing voices, paired with the browser's native Speech Synthesis API for lyric vocalization.

Zero dependencies. Single HTML file. Instant music.

---

✨ Key Features

Feature	Description
🎹 6 Genres	Pop, Rock, Jazz, Electronic, Classical, Lo-fi — each with unique scales, chords, and drum patterns
🎤 Dual Vocal Engine	Formant synthesizer for melodic "aah/ooh" vocals + Speech Synthesis for lyric articulation
🎼 Text-to-Melody	Character-to-note mapping algorithm that generates scale-appropriate melodies from any text
📊 Live Visualizer	Real-time 80-bar frequency spectrum analyzer with genre-colored gradients
🇮🇩 Bilingual Presets	5 original Indonesian songs with English translations
🎚️ Full Controls	Genre, key, tempo (60–180 BPM), volume, play/pause/stop
📱 Responsive	Works on desktop and mobile browsers
⚡ Zero Dependencies	Single index.html file — no npm, no build step, no server

---

🚀 Quick Start

Option 1: Live Demo

Visit sonic-ai-dun.vercel.app — no installation needed.

Option 2: Run Locally

git clone https://github.com/romizone/sonic-ai.git
cd sonic-ai
open index.html

Option 3: Local Server

python3 -m http.server 3000
# Open http://localhost:3000

---

🎵 How It Works

┌─────────────┐     ┌──────────────────┐     ┌─────────────┐
│  Input Text  │────▶│  Text-to-Melody  │────▶│   Melody    │──┐
│  / Lyrics    │     │  (char → note)   │     │  Oscillator │  │
└─────────────┘     └──────────────────┘     └─────────────┘  │
                                                               │
┌─────────────┐     ┌──────────────────┐     ┌─────────────┐  │  ┌──────────┐
│ Genre Config │────▶│  Chord Engine    │────▶│  Pad Synth  │──┼─▶│  Master  │
│ (scale,bpm) │     │  (I-IV-V-I etc)  │     │  + Bass     │  │  │  Gain    │
└─────────────┘     └──────────────────┘     └─────────────┘  │  │          │
                                                               │  │    ┌─────┤
┌─────────────┐     ┌──────────────────┐     ┌─────────────┐  │  │    │Reverb│
│  Formant    │────▶│  Bandpass Filter  │────▶│  Vocal      │──┤  │    └──┬──┘
│  Vocal Syn  │     │  Chain (F1,F2,F3) │     │  "Aah/Ooh"  │  │  │       │
└─────────────┘     └──────────────────┘     └─────────────┘  │  ├───────┘
                                                               │  │
┌─────────────┐     ┌──────────────────┐     ┌─────────────┐  │  │  ┌──────────┐
│  Drum       │────▶│  Kick/Snare/HH   │────▶│  Drum Bus   │──┘  └─▶│ Analyser │
│  Pattern    │     │  Synthesis       │     │             │      │ + Output │
└─────────────┘     └──────────────────┘     └─────────────┘      └──────────┘

---

🎤 Preset Songs

#	Song	Artist	Genre	BPM	Key
1	Senja di Sudirman	Dian Sastro	Pop	110	C
2	Hujan di Senopati	Wulan	Jazz	95	Am
3	Kereta Terakhir	Davina	Rock	125	Em
4	Kopi dan Janji	Titi Kamal	Lo-fi	85	F
5	Lampu Kota	Davina	Electronic	128	G

All songs feature Jakarta city themes with both Indonesian and English lyrics.

---

🏗️ Architecture

Audio Synthesis Engine

Layer	Technology	Purpose
Melody	OscillatorNode + LowpassFilter	Genre-specific waveforms with vibrato
Vocals	3x Detuned Oscillators + BandpassFilter chain	Formant synthesis (vowel-like "aah/ooh")
Speech	SpeechSynthesisUtterance	Lyric articulation with pitch/rate tuning
Chords	Detuned OscillatorNode pairs + LowpassFilter	Pad sounds with smooth crossfade envelopes
Bass	OscillatorNode + LowpassFilter (400Hz)	Genre-specific waveform bass lines
Drums	OscillatorNode + AudioBuffer (noise)	Synthesized kick, snare, hi-hat
Reverb	ConvolverNode (procedural impulse)	Exponential decay with early reflections
Visualizer	AnalyserNode + Canvas 2D	80-bar frequency spectrum at 2x resolution

Genre Configurations

Each genre defines a unique combination of:

• Scale — Ionian, Minor, Blues, Pentatonic, Dorian
• Chord Progression — I-IV-V-I, i-iv-III-i, Imaj7-IVmaj7, etc.
• Waveforms — sine, triangle, sawtooth, square
• Formant Frequencies — F1/F2/F3 tuning for vocal character
• Drum Pattern — Beat placement and swing feel
• Effects — Reverb decay (1.2s–4.0s), filter cutoff, wet/dry mix

---

📊 Comparison with Deep Learning

	SonicAI	Suno AI
Voice Quality	Formant synth + TTS	Neural vocal synthesis
Latency	Instant (client-side)	Seconds–minutes
Dependencies	None (browser only)	GPU servers
Offline	Full support	Requires internet
Cost	Free & open source	Subscription
Privacy	100% local	Data sent to servers
File Size	~40KB	Multi-GB models

---

📄 Academic Paper

The full technical paper is available at SonicAI_Paper.pdf, covering:

• System architecture and audio signal flow
• Text-to-melody conversion algorithm
• Formant vocal synthesis with bandpass filter chains
• Genre-adaptive chord progression and drum pattern engines
• Chrome autoplay policy compliance
• Comparison with deep learning approaches
• Limitations and future work

---

🛠️ Tech Stack

---

📝 License

This project is open source and available under the MIT License.

---

👤 Author

Romi Nur Ismanto

• Email: rominur@gmail.com
• GitHub: @romizone

---

_{Built with Web Audio API | Deployed on Vercel | Made with ❤️}