Stem Separator (Mel-Band RoFormer + SCNet)
Free.ai (self-hosted)
·
audio_separate
·
~1000 Ama-token ngalinye track
Stem Separator (Mel-Band RoFormer + SCNet) yi an imodeli yokuhlukaniswa kwe-stem. Onamandla kakhulu ku-Vocal/stem separation. I-self-hosted ku-Free.ai GPUs - isebenza ngokukhululekile ngaphezu kwe-token pool yakho yansuku zonke (1,000 tokens Ingoma ngayinye). Kukhishwa ngaphansi kwe-MIT + Apache-2.0 - ukusetshenziswa kokuthengiswayo kuvunyelwe ku-Free.ai.
Sebenzisa nge-API
I-REST API ehambisana ne-OpenAI. Dala isithonjana bese ubiza le modeli emaminithini.
curl -X POST https://api.free.ai/v1/music/separate/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"demucs","prompt":"your prompt here"}'Imibuzo ebuzwa kaningi
Stem Separator (Mel-Band RoFormer + SCNet) ihlukanisa umsindo oxhunywe ku-stems - vocals, drums, bass, nezinye izisetshenziswa - ngayinye njengefayela le-audio. Layisha i-MP3 noma i-WAV futhi Stem Separator (Mel-Band RoFormer + SCNet) ibuyisela amafayela we-stem ahlukene angu-4 ongawafaka kabusha noma ungayicishi.
Kuhle kakhulu kulezi zingoma ezixhunywe kahle: ama-vocals nama-instrumentals ahlukaniswe nge-bleed encane, nama-drums nama-bass aphuma ahlanzeke. Ama-records adala, aphilayo noma acindezelwe kakhulu ahlukaniswe nge-crosstalk engaphezulu phakathi kwe-stems.
Izinhlamvu ezine: ama-vocals, ama-drums, ama-bass nezinye. I-instrumental track (noma yini ngaphandle kwama-vocals) iyatholakala njenge-output ecindezelwe kabili.
MP3, WAV, FLAC, M4A, OGG kuze kube yi-50 MB. Ukungeniswa kwe-stereo kunikeza izimpendulo ezihlanzekile; i-mono ikhuphuka ibe yi-stereo ngaphambi kokusebenza.
Stem Separator (Mel-Band RoFormer + SCNet) ihamba ku-GPUs yethu ephikisana ne-pool yakho yamahhala yansuku zonke kuqala; $5 → 200,000 i-token ekhokhelwa. Ngokungafani ~1,000 i-token ngomzuzu womsindo.
Yebo - khipha izitho, bese ulanda "i-drums + bass + ezinye" ezixhunywe ndawonye ukuze usebenzise i-instrumental, noma ulanda "i-vocals" kuphela ukuze usebenzise i-a-cappella. Ifomu elingezansi line-click-one "instrumental" output.
Yebo - ukuhlukaniswa akushintshi isikhathi noma i-pitch. Izinhlamvu zilinganise ifreyimu-nge-freyimu nomthombo ukuze ziphinde zixhunywe ngokuhlanzekile.
Yebo - /batch/ ithatha isibaya sefayela lomsindo. Imiphumela yonke (izinhlamvu ezi-4 ngasinye) iwela /account/?tab=history. Ukugcina isibaya-mhlahlandlela sebenzisa i-API.
Yebo. UTHENGA umsindo wakho ku /v1/music/separate/ futhi uthola i-URL eyodwa yokulayisha phezulu ngayinye ye-stem emuva. /api/ ine-reference ephelele, kufaka phakathi imodeli id ethunyelwe.
Udinga ilungelo lomthombo womsindo (u Recording wakho, i-licensed material, noma i-song ene-explicit remix permission). Free.ai angakwazi ukunika ilungelo le-material ongayinikazi; ukuhlukanisa i-chart pop song akuvumelekile ukushicilela izitho.
WAV ngokuzenzakalela (akukho phutha, ukulungele ukuxhuma). MP3 itholakala kufomethi yokukhetha ukukhishwa okuncane.
Isikhathi sangempela somphumela yi 0.1-0.3× - umculo wemizuzu engu-4 uhlukaniswa ngamasegundo angu-30-80. Sebenzisa inkinobho yokulinganisa ukuvala i-tab.