VoiceForge Karaoke — JSON App Summary

🔗 Source

Where it lives

App Page

producingtechnology.com/65-apps/taoyifei_182158_15200435_yt684 (3).html

SRC_URL

producingtechnology.com/a1/15-json/taoyifei_182158_14904036_yt684.json

🎙️ App Behavior

An AI karaoke dashboard

VoiceForge Karaoke presents itself as an AI-powered living-room karaoke app. The landing screen shows a pulsing mic logo, a gradient-text header, and an ambient pink/cyan/purple glow behind everything. On load, the app fetches a session payload and renders a two-column dashboard — a big "Now Playing" panel on the left and a sidebar with queue, leaderboard, and AI feature toggles on the right.

🏠 Living Room mode ⚡ Crowd Energy bar 🎹 Key & BPM 👑 Legendary difficulty

Now Playing

Big gradient "album art" card, song title in Righteous display font, artist name in pink, meta chips for key/BPM/duration/difficulty, and three AI toggles below (pitch assist %, harmony voices, style transfer).

Lyrics Display

One line at a time shown in a dark box, with the target notes listed underneath (e.g. Bb3 · Bb3 · C4 · Bb3 · G3 · F3).

Playback Controls

Five circular buttons: prev lyric, prev track, play/pause, next track, next lyric. Pressing play auto-advances the lyric line every 3 seconds.

Crowd Reactions

Five reaction buttons (cheer, clap, gasp, sing along, lighter wave). Clicking spawns a burst of floating emoji that drift up the screen, and bumps the crowd-energy bar by 5%.

Up Next / Leaderboard / AI Features

Sidebar cards showing the song queue (click to jump), a two-person leaderboard with avatar emoji and badges, and three AI model tiles with latency readouts you can click to disable/enable.

⚠️ Didn't Work As Expected

What's off

Pressing Play plays nothing

The app has no audio at all — the play button just starts a setInterval that advances the lyric every 3 seconds.

Critical

For a karaoke app the absence of audio is the defining issue. No backing track, no vocals, no metronome — just silent lyric paging. Even a web-MIDI beep at each note would communicate "this is a demo of the UI, audio lives elsewhere."

Lyric timing ignores the data

Each lyric in the JSON has real start_ms and end_ms fields, but the app just advances every 3000ms.

High

"Bohemian Rhapsody" line one runs 48.0–55.0s (a 7-second line), and line two runs 55.5–62.0s (6.5s). Fixed 3s advancement means lyrics desync immediately. The timing data is right there in the payload, unused.

Settings read-only despite looking interactive

Pitch assist shows "(40%)", harmonies shows "(2)", but there's no way to change either number.

High

The toggles flip the boolean on/off, but pitch_assist_strength (0.4) and harmony_voices (2) are displayed inline as if they were controls. No slider, no plus/minus, no dropdown. Same with style_transfer — the chip appears when a style is set, but the JSON's available_styles array of six options is never offered as a picker.

Leaderboard never updates

Scores are whatever the JSON says, forever.

Medium

Mashing the clap button 50 times can push crowd energy to 100% but nobody's score changes. score_total, songs_sung, best_streak are all static readouts rather than live state.

Rich singer data never surfaced

Each singer has a vocal range, song count, and streak; the UI only shows score and two badges.

Medium

vocal_range: { lowest_note: "A3", highest_note: "C6" } is in the payload but nowhere on screen. A karaoke app deciding song suggestions by vocal range would be obvious — this doesn't.

Reaction spam is unbounded

Each click spawns 5 floating emoji with no cooldown and no cap.

Low

Rapid button-mashing creates dozens of simultaneous floating emoji and pushes crowd energy to 100% in seconds with no indication it's maxed. Also, once it hits 1.0 it just stays there silently.

Queue is read-only

You can jump to a song but not add, remove, or reorder.

Low

queued_by is shown on each song, which implies multiple people queue songs, but there's no add-to-queue UI and no way to see which singers exist beyond the two in singers[].

🚀 Improvement Prompt

Best attempt at a fix-it prompt

Rework this single-file VoiceForge Karaoke app so the AI-feature UI actually does something and the rich fields in the JSON are no longer wasted. Keep the neon pink/cyan/purple aesthetic, the Righteous display font, the gradient top-bar on each card, and the floating emoji reactions — only change or add what's listed below.

Drive lyrics from the actual timing data

Replace the hardcoded 3-second interval with a real clock. When play is pressed, track elapsed ms since start and switch the current lyric based on each line's start_ms / end_ms. Show a thin gradient progress bar under the lyric box fed by the line's remaining time.

Make the AI settings editable

Turn the "Pitch Assist (40%)" chip into a chip-plus-slider that adjusts pitch_assist_strength from 0–100%. Turn "Harmonies (2)" into a stepper with +/- buttons bounded at 0–4. Replace the static style-transfer chip with a dropdown seeded from ai_models.style_transfer.available_styles.

Add a mock audio layer using the Web Audio API

No backing track needed — just a sine-wave oscillator that plays the current lyric's notes in sequence when playing. The note letters already exist in each lyric's notes array. This turns the app from a silent demo into something that actually responds to play/pause and AI toggles (e.g. pitch correction snaps the oscillator to the nearest semitone).

Make the leaderboard live

Add a "Performance Score" that ticks up while playing based on crowd energy and the AI settings (e.g. +1/sec baseline, ×1.5 if harmonies are on, ×2 during legendary songs). Update score_total in state when a song ends and re-sort the leaderboard with a smooth reorder animation.

Show each singer's vocal range

Add a small "Singers" card above the leaderboard showing each singer as an avatar chip with their vocal range rendered as a mini horizontal bar from C3 to C6 and their best_streak as a flame count. Highlight any queued song whose root note falls inside the current singer's range.

Tame the reaction spam

Cap floating emoji at 20 simultaneous elements; apply a 200ms cooldown per reaction button with a radial cooldown ring animation. When crowd energy hits 100%, briefly pulse the energy bar gold and flash a "MAX HYPE" ribbon across the top of the Now Playing card.

Make the queue editable

Add an "Add to queue" button at the bottom of the sidebar that opens a small modal with title, artist, difficulty, and "queued by" (dropdown of existing singers). Allow drag-to-reorder and swipe-left-to-remove on queue items. New queue entries should get sensible defaults for ai_settings.

Single self-contained HTML file, graceful degradation

No build step, no external libs beyond the existing Google Fonts import. If any JSON field is missing (no vocal_range, no available_styles, etc.), the corresponding UI should just not render rather than throwing.