deezer

@itsnoahd i also saw this before deezer/spleeter but this is only split music and vocal and its not that powerful yet

ステム分離してからマッシュアップする：短い制作メモ ステム分離してからマッシュアップする：短い制作メモ 仮BGMや短いデモで「A曲の歌い回しは残したいが、B曲のビートの方が映像に合う」という詰まり方があります。フルミックス同士をいきなり重ねると、どちらも主張してVOとぶつかりやすいです。今回のメモは 先にステム分離 → 必要なパートだけ残す → 必要ならマッシュアップ という短い手順です。 作業場所は Doremi。分離は AIステム分離、2曲を新しい1曲にまとめるときは AIマッシュアップメーカー を使います。 なぜ「分離を先」にするか 音源分離は研究・実装の両面で長く扱われてきた問題で、Deezer のオープン実装 Spleeter が実用的な出発点として知られています。製品UIでも本質は同じで、「ミュートしたいパートを単独で扱えるか」が採否基準になります。 マッシュアップはもともと2曲以上の素材を新しい1曲にまとめる手法です（概要: Mashup (music)）。入力がフルミックスのままだと、不要なボーカルやシンセまで一緒に引っ張られやすいので、短い映像用途では分離を前段に置くと失敗が減ります。 手順（最短） キーパーを2本決める（雰囲気が近いもの。完璧さは後回し）。 それぞれをステム分離（用途に応じて 2 / 4 / 6 / 10）。 Webプレイヤーでソロ/ミュート確認（VO下で聴く。ヘッドホンソロだけだと密度を誤判定しやすい）。 必要なステムだけ書き出し（全部保存しない）。 まだ「1本の聴感」が欲しければマッシュアップ（曲調・テンポ感・ボーカル有無を一文で固定）。 トラック数の目安 目的 目安 カラオケ / インスト化 2（ボーカル / 其余） ドラム差し替え前提のリミックス 4 特定楽器の耳コピ・練習 6〜10 実務メモ 1パス1意図。分離とマッシュアップを同時に欲張ると、どちらが効いたか分からなくなります。 候補は2〜3本で打ち切る。 採用した「曲調一文」はリポジトリ横に残す（次回の映像尺で再利用）。 完璧な楽曲制作ではなく、レビュー用の「尺と密度が合う1本」を、フルミックス衝突なしで収束させるためのメモです。同じ前段分離の指示テンプレがあれば交換したいです。

Separation Mode as a Pipeline Choice: 2-Stem vs 4-Stem Before You Lengthen If your short-form pipeline treats “split the mix” as a single boolean, you will over-export stems and under-learn from listens. Separation mode is a design choice: it decides which axes you can mute later and how much review tax you pay. This note frames 2-stem vs 4-stem as an API-shaped decision inside WAV AI, then shows where lengthening belongs—after the stem graph is stable. !Neural demixing feature art Background: why mode exists at all Music source separation is not one algorithm with one output shape. Research stacks differ in how they carve a stereo mixture—Deezer’s Spleeter popularized practical multi-stem demos; Meta’s Demucs line explores hybrid spectrogram/waveform models. Downstream tools expose a smaller product surface: modes like vocal/instrumental vs vocals/drums/bass/other. Your job as a pipeline author is to pick the minimum stem set that unlocks the edit, not the maximum stem set the model can emit. Mode as an interface contract Think of the splitter like an endpoint: | Mode | Rough contract | Good for | Bad for | |------|----------------|----------|---------| | 2-stem | mixture → {vocals, instrumental} | Karaoke beds, VO under instrumental, quick mute of lead | Moving kick vs bass independently | | 4-stem | mixture → {vocals, drums, bass, other} | Sample flips, DJ bootleg prep, drum replacement | Reviewers who will not solo four faders | Same upload, different review graphs. If your checklist only says “run stem split,” reviewers will thrash. A minimal pipeline sketch brief → choose_mode(2|4) → split → solo_gate → export_needed → (optional) rebuild_in_daw → extend_if_duration_gap → ship_max_2_candidates choose_mode - Need only “voice off” or “voice alone”? → 2-stem. - Need drums without bass mud, or bass without kick bleed? → 4-stem. - Unsure? Start 2-stem. Escalate only when the solo gate fails for a layered reason. solo_gate Reject before export when: - Hats smear into vocals (or the reverse) enough to hear on phone speakers. - The instrumental still carries ghost syllables under a new VO. - Transients on drums feel soft enough that the edit will fight the picture cuts. The AI Stem Splitter UI’s mute/solo loop is the gate—not a folder dump of every stem “for later.” Where lengthening sits Lengthening is a time-axis operation. It assumes identity (key, groove, vocal tone) is already acceptable. Call an AI Music Extender only after: 1. The stem graph (or original mix) passed solo_gate. 2. You marked a continuation timestamp after the last good bar. 3. You know whether the continuation should stay instrumental (common under VO) or carry lyrics. Extending before mode selection just freezes a bad stem graph into a longer file. Extending after a failed solo_gate teaches the wrong lesson (“AI outro bad”) when the real bug was bleed. Practical defaults for small teams - Default mode: 2-stem for talk-over and social beds. - Escalate to 4-stem: when the brief names a drum or bass problem explicitly. - Export policy: only stems referenced by the timeline ticket. - Candidate cap: two continuations per cut; log one-line failure reasons. What this note is not It is not a bake-off of model architectures, and it does not cite product marketing percentages. Use your own listens and your own picture lengths as ground truth. Mode first, length second—treat both as typed steps in the pipeline, not as interchangeable “AI music” buttons.

Stem count as an API choice: 2 vs 4 vs 6+ for short remix pipelines When I wire AI audio into a small content pipeline, the expensive mistake is treating “stem split” as one opaque button. Track count is closer to an API parameter: too few and you can’t mute the part that fights VO; too many and you babysit bleed and file juggling. This note is how I pick 2 / 4 / 6 / 10 before I touch arrangement. Workspace: Doremi. Separation lives in the AI Stem Splitter; mashups come later only if I still want a fused listen. Why stem count matters more than model marketing Source separation is a researched problem—Deezer’s Spleeter made a practical baseline famous by shipping vocals/drums/bass/other (and finer recipes) as an open tool. Product UIs vary, but the acceptance criteria stay the same: can I solo the part I need without wrecking the rest? My default mapping for short-form video beds: | Need | Start with | Notes | |------|------------|-------| | Karaoke / VO-safe instrumental | 2 stems | Vocal vs rest; fastest listen | | Remix with drum swap | 4 stems | Vocals / drums / bass / other | | Practice or instrument-focused edit | 6–10 | Guitar/piano isolation when it matters | | “I might mash two songs later” | Split both sources first | Don’t mash until parts pass mute tests | Pipeline sketch (builder view) 1. Import the keeper mix (MP3/WAV/FLAC—whatever you already export). 2. Choose the minimum stem count that unblocks the next edit. 3. Solo/mute in the web player; reject splits with obvious vocal bleed under speech. 4. Export only the stems you’ll actually drop into the DAW. 5. Optional: if two cleaned sources still want a hybrid listen, run them through the AI Music Mashup Maker with a tight style prompt (“same BPM feel, no new vocal hook, bed density”). Mashups themselves are an old craft of combining two recognizable songs into one new piece (Mashup (music)); stems just give you cleaner inputs than slamming two full mixes together. Failure modes I budget for - Over-splitting: 10 tracks for a 20s UI tour. You won’t use them; you’ll lose an evening naming files. - Under-splitting: 2 stems when the problem is a bright synth, not the vocal. You mute the wrong layer. - Mashup-before-split: fusing two muddy mixes, then wishing you could mute one vocal. Split first when control matters. Ticket-style prompt scraps stems: 4 keep: drums+bass pulse under VO export: instrumental + vocal isolated forbid: new arrangement yet — separation only This isn’t a DSP paper—just the shortest acceptance loop I’ve found so stem jobs don’t turn into another free-BGM scavenger hunt. If you version stem recipes next to your video repo, drop a schema; I’ll copy the good parts.