EPonGO Experimental Staging

Work-in-progress features. Share links, listen to outputs, evaluate quality before we promote anything to production.

1 · Suno Stems + WAV Master

For Suno-generated songs. Pulls 12 individual instrument stems (WAV) + the uncompressed master WAV directly from Suno. Runs on suno.epongo.com.

LIVE — API endpoint deployed on suno.epongo.com. Costs 13 Suno credits per song.

Details to Follow
Robert Daly · Song Lab generated
Suno clip 5aea4288-46ba-4aba-83ba-16a90e5bd007 · 3:13 · alternative metal
Master (WAV, 35.4 MB)
12 stems (WAV each)
Vocals Backing Vocals* Drums* Bass* Guitar* Keyboard* Percussion* Strings* Synth* FX* Brass* Woodwinds*

* MP3 (older pull, before WAV-per-stem was default)

Don't Get It Twisted (Live at Cheyenne Frontier Days)
Sadie Night · Suno-generated, NOT her real recording
Suno clip 58f3c9e1-9068-45e2-a4ee-7498f5e2c55b · 3:27 · country · derived from prompt "make live version of this song"
Heads up: Suno cannot actually transform existing audio. When we prompted it with "make a live version of this song," it generated a brand-new song from scratch that shares the title but has different melody/lyrics/vocalist. The stems and karaoke below are technically clean — but they're of the Suno-generated song, not Sadie's real recording. For Sadie's actual song, wait for the Demucs pipeline (below).
Master (WAV)
12 stems (WAV each)
Vocals Backing Vocals Drums Bass Guitar Keyboard Percussion Strings Synth FX Brass Woodwinds

2 · Karaoke Mix (no lead vocals)

Mixes all stems except the lead vocals into a single track. WAV lossless + MP3 for streaming. Keeps backing vocals as harmonies by default.

LIVE — API endpoint on suno.epongo.com. Runs ~30s once stems exist. No Suno credits burned.

Details to Follow — Karaoke
Robert Daly
Karaoke (MP3, 7.7 MB)
Download WAV Download MP3
Don't Get It Twisted (Live) — Karaoke
Sadie Night · Suno-generated variant
Karaoke (MP3)
Download WAV Download MP3

3 · Chord Lead Sheet

Auto-generates a lead sheet with lyrics + chords + key + tempo. Runs on any audio (uploaded WAV/MP3 or Suno-generated).

QUALITY NOT ACCEPTABLE — chord jitter is unusable, lyrics ~50% word accuracy. Rebuild in progress: run Whisper-large on isolated vocals stem, run neural chord recognizer on isolated bass+other stems.

Don't Get It Twisted — Lead Sheet (v1, unusable)
Sadie Night · uploaded WAV
Detected: G major · 172 BPM · 190 seconds · 20 chords detected (should be ~6) · 112 words transcribed (probably 250+)
Open HTML lead sheet Raw JSON

This v1 output is here so we can compare against v2 quality once the rebuild is done. Don't use for real work.

4 · Demucs stems (for uploaded songs)

Works on any uploaded audio — MP3, WAV, M4A. Facebook's open-source source-separation model. Outputs 4 stems: vocals, drums, bass, other. This is what we need to make stems / karaoke / lead sheets on Sadie's actual recording (and any other uploaded track).

INSTALLING NOW — CPU-only PyTorch + demucs on Vultr. First test coming after install completes. Expected runtime per 3-min song: 15-30 min.

Don't Get It Twisted (regular version)
Sadie Night · uploaded WAV · queued
Once Demucs is installed we'll process this first. Deliverables: 4 WAV stems + karaoke mix + retry the chord sheet with the cleaned vocals.

How to trigger new extractions

The suno.epongo.com API accepts triggered extractions. Auth via shared secret.

curl -X POST https://suno.epongo.com/extract \
  -H "X-Extract-Token: $SUNO_TOOLS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"clip_id":"<suno-clip-uuid>","r2_prefix":"song/<track-uuid>","trigger":true,"wav_stems":true}'

# Poll:
curl -H "X-Extract-Token: $SUNO_TOOLS_TOKEN" \
  https://suno.epongo.com/jobs/<job_id>

# Karaoke on existing stems:
curl -X POST https://suno.epongo.com/karaoke \
  -H "X-Extract-Token: $SUNO_TOOLS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"r2_prefix":"song/<track-uuid>","stems_format":"wav"}'