Spleeter
Free.ai (self-hosted)
·
audio_separate
·
~1000 tokens per track
Spleeter ke a Mofuta oa ho tlohela stem e hahiloeng ke Deezer. E ka ba e le DJ/production workflows, karaoke extraction.. E na le li-Free.ai GPUs tse nang le li-server tse 100 tse nang le li-server tse 100 tse nang le li-server tse 100. E lokollotsoe ka MIT — ho sebelisoa ka khoebo ho lumelloa ka Free.ai.
Ho sebelisa ka API
REST API e lumellanang le OpenAI. E etsa konopo'me e bitsa mofuta ona ka metsotsoana.
curl -X POST https://api.free.ai/v1/audio/separate/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"spleeter","prompt":"your prompt here"}'Lipotso tse botsoang khafetsa
Spleeter e arola lipina tsa audio tse tšehetsoeng ho li-stems — li-vocals, li-drums, li-bass, le li-instruments tse ling — e 'ngoe e le faele ea audio ea eona. Kopitsa MP3 kapa WAV' me Spleeter e tla u khutlisa li-stemfiles tse 4 tse fapaneng tseo u ka li remix kapa u li hlakole.
Demucs-family separators are state-of-the-art — vocals are nearly studio-quality on modern productions, drums and bass are very clean. Old or heavily-compressed source material yields lossier separation.
Standard 4-stem: vocals, drums, bass, tse ling. Litšoantšo tse ling li fana ka 6-stem (ho eketsa piano + guitar ka boeona) — khetha mofuta ka har'a foromo.
MP3, WAV, FLAC, M4A, OGG ho fihlela ho 50 MB. Stereo input e fana ka litlamorao tse hloekileng; mono e thehiloe ho stereo pele e phekoloa.
Spleeter e sebetsa ka GPUs ea rona ho ea ka pool ea hau ea mahala ea letsatsi le letsatsi pele; $ 5 → 200,000 tokens tse tšehelitsoeng. Ka li-tokens tse ka bang 1,000 ka metsotsoana ea audio.
E-na — tlosa li-stems, ebe u kenya "drums + bass + other" tse hokahantsoeng bakeng sa li-instrumental, kapa u kenya "vocals" feela bakeng sa a-cappella. Form e ka holimo e na le "instrumental" output e ka ntlafatsoang ka ho tobetsa feela.
Ee — ho arola ha ho fetole nako kapa palo. Li-stems li hokahanya li-frame-by-frame le sehlooho ka hona li remix ka mokhoa o hloekileng.
Ee — /batch/ e amohela lebokose la lifaele tsa audio. Lihlahisoa tsohle (li-stems tse 4 ka mohloli) li tla ka /account/?tab=history. Ho boloka moqhaka oa lebokose la lifaele sebelisa API.
E-na — POST audio ea hau ho /v1/music/separate/ le model="Spleeter". E khutlisa URLs tse 4 tsa ho kenya (e 'ngoe ka stem). /api/ e na le ho ngolisoa ka botlalo.
U hloka litokelo tsa audio ea mohloli (ho ngolisoa ha hau, thepa e nang le laesense, kapa ngoana ea nang le tumello ea ho e fetola ka ho hlakileng). Free.ai ha e ka u fa litokelo tsa thepa eo u sa e nang le eona; ho tlohela ngoana oa pop oa chart ha ho lumellane ho phatlalatsa stems.
WAV ka ho laoloa (ha ho na liphoso, e loketse ho ntlafatsoa). MP3 e fumaneha ho khethang li-format bakeng sa ho jarolla ka tlase.
Nako ea 'nete ke 0.1–0.3× — ngoana ea metsotsoana e 4 o tlohela ka metsotsoana e 30–80. Sebelisa konopo ea ho lekola ho koala tab.