TRK:IQ — Timing Tracks

TRK:IQ listens to your song and pulls out everything a sequencer needs: it splits the audio into separate instruments, finds every drum hit and beat, and syncs the lyrics down to individual mouth shapes for singing faces. Everything lands in one .xtiming file.

Under the hood, three engines run in a relay:

• Demucs separates the song into six stems — vocals, drums, bass, guitar, piano, and other.
• Essentia finds the BPM, beats, downbeats, and every hit on every stem.
• The lyrics aligner syncs your lyrics to the vocal stem — phrases, words, and phoneme mouth shapes.

Step 1: Upload

Drop in an MP3, WAV, FLAC, or M4A. TRK:IQ reads the file's metadata and goes hunting for lyrics on its own.

Lyrics: three ways in

Lyrics power the singing-face tracks. The more timing info the lyrics carry, the sharper the final sync. Pick whichever path fits:

1. Auto-fetch (do nothing). TRK:IQ checks LRCLIB using your file's artist and title. If it finds synced lyrics — lines with [mm:ss.xx] timestamps — alignment confidence jumps about 10–20%, and a "synced" badge appears.
2. Paste plain lyrics. If auto-fetch misses, paste the words yourself. The AI aligner figures out the timing from scratch.
3. Paste synced lyrics (LRC format). If you have timestamped lyrics, paste them and TRK:IQ uses the timestamps as anchors. Each line needs a [mm:ss.xx] tag up front, with at least 3 valid lines:

   [00:15.20] I'm on the edge of glory
   [00:19.80] And I'm hanging on a moment of truth

Step 2: Processing

Click Generate and a GPU server takes it from there. A progress ring and step pills track five phases:

If the GPU is busy you'll see "Waiting for our turn…" with a queue position. Waits are usually a few seconds — just enough time to untangle an extension cord.

Step 3: Export

The stats banner shows the detected BPM, total tracks, words (if lyrics ran), and total timing marks. Below it, every track sits in a list organized by tabs — All, Drums, Melodic, Structure, Singing Faces. Each row shows:

• A checkbox — only checked tracks go into your download. Re-check and download again any time, no reprocessing.
• A source badge — green AI (high confidence ≥70%), amber AI (worth a look), Synced (built from LRC timestamps), or Local/Est. (fallback estimates).
• The confidence range and hit count ("342 hits", "156 words", "480 phonemes").

What's in the file

One .xtiming, up to 14+ tracks: five drum tracks (kick, snare, hi-hat, toms, cymbals), onset tracks for bass, guitar, piano, vocals, and "other," a labeled Beat Grid, a Song Structure track (intro / verse / chorus / bridge / outro), and the Lyrics (Lead) track — three stacked layers of phrases, words, and Preston Blair phoneme codes, which is exactly the format xLights singing faces expect.

Importing into xLights

1. 1

   Open your sequence

   Load (or create) the sequence for this song. Make sure its audio file matches the one you gave TRK:IQ.
2. 2

   Import the timing track

   In the sequencer, right-click the Timing Tracks area → Import Timing Track → pick your .xtiming. xLights lists every track inside — import all or cherry-pick — then click OK.
3. 3

   Check what arrived

   Onset tracks show thin markers at each hit. Structure tracks show labeled sections. The Lyrics track expands into its three layers: phrases, WORDS, and phoneme codes.
4. 4

   Wire up singing faces

   Select your face model and point its singing-face effect at the lyrics track. The phoneme codes (AI, O, E, U, MBP, FV…) are the standard set xLights expects — no conversion needed.
5. 5

   Sequence away

   Flash arches on every kick. Chase the beat grid. Go subtle on verses and big on the chorus. This is the fun part.

FAQ