Step Audio 3 Uyasebenza

I-StepAudio 3 iwuchungechunge lwamamodeli amakhulu ezwi olwethulwe yi-Jieyue Xingchen, oluhlanganisa amamodeli amahlanu: i-Realtime, ASR, TTS, Gen ne-Music. Phakathi kwawo, ukunemba kokucabanga ngezwi kwe-Realtime kungu-99.7%, kanti izinga lamaphutha amagama le-ASR lingu-1.7%; womabili akleliswe endaweni yokuqala emhlabeni ku-Artificial Analysis. I-StepAudio 3 iyakwazi ukuzwa nenkuluma, futhi iqonde imizwa nethoni, yenze ukucabanga okuyinkimbinkimbi, ikhiqize izwi nemisindo ndawonye, futhi ibambe iqhaza kulo lonke uhlelo lokwakha umculo. Ihlanganisa izimo ezifana nezingxoxo zesikhathi sangempela, ukuguqula amazwi abe imibhalo engezansi, ukudabula amazwi kumafilimu nakumabonakude, nokukhiqiza umculo.

I-Realtime, ingxoxo yesikhathi sangempela: isekela ukuxhumana kwemvelo kwezindlela zombili, iyakwazi ukuzwa, icabange, yahlulele ijubane lengxoxo futhi yenze imisebenzi.
I-ASR, ukuqaphela inkulumo: iguqula ngokuqondile isiShayina, isiNgisi, izilimi zesifunda kanye nenkulumo exubile yesiShayina nesiNgisi ibe umbhalo, iqonde amagama obuchwepheshe nezimo eziyinkimbinkimbi.
I-TTS, ukuhlanganisa inkulumo: ikhiqiza izwi elifana nelomuntu wangempela, iphinde ikhiqize ithoni, imizwa, izikhawu kanye nemininingwane efana nokuhleka nokungabaza.
IGen, ukukhiqiza umsindo: ikhiqiza izwi lomuntu, imisindo, umsindo wemvelo nomculo wangemuva ngesikhathi esisodwa, futhi ilawule ngokunembile isikhathi sento ngayinye.
IMusic, ukudala umculo: isekela ukudala kusukela ekuqaleni, ukuphelezelwa kokucula ngaphandle komculo, ukucula kabusha izingoma kanye nokudala ngokuxoxisana okuningi kusetshenziswa indlela yokubhala amanothi ye-ABC.

I-Realtime: isebenzisa isakhiwo semvelo sezindlela zombili, lapho ukulalela nokukhuluma kwenzeka ngokuhambisana; phakathi kwengxoxo eqhubekayo, imodeli iqonda umsindo, yenze ukucabanga futhi ikhiqize inkulumo ngesikhathi esisodwa, isekele ukucabanga nokukhiqiza okuhambisanayo kanye nokwenziwa kwe-Tool Call ngokungavumelani.
I-ASR: ihlanganisa ukuqaphela inkulumo okunembile nemodeli yolimi enkulu, isebenzise ukuqonda komongo, ulwazi namandla okucabanga emodelini ukulungisa amagama anemisindo efanayo, amagama abantu, amagama ezindawo namagama obuchwepheshe, isuke “ekuguquleni umsindo ube umbhalo” iye “ekuqondeni umongo”.
I-TTS: isebenzisa isakhiwo sokukhiqiza okusakazwayo ukuze ikhiqize futhi idlale ngesikhathi esisodwa, ihlangabezane nezidingo zokubambezeleka okuphansi ekuxhumaneni kwesikhathi sangempela; iphinde imodeli umbala wezwi, iphimbo, isigqi nezikhawu zokuphefumula, ilungise ngokuzenzakalelayo imizwa nethoni ngokusekelwe encazelweni.
IGen: ithatha imodeli eyodwa esikhundleni kwezinqubo ezihlukene zendabuko zokudabula amazwi, imisindo nemiculo, ikhiqize izingqimba eziningi zemisindo ngesikhathi esisodwa ngokusekelwe encazelweni yolimi lwemvelo nomsindo oyisithenjwa, futhi ibambe iqhaza ekwakhiweni komsindo ngokuhlela isikhathi kwengxoxo, imisindo nomsindo wemvelo.
IMusic: isekela kokubili ukukhiqiza kusukela ekuqaleni nokudala okunemibandela; ngokuqonda izici zomculo ekuculeni ngaphandle komculo, emazwini engoma nasemculweni, inikeza ukulawula okuhlelekile kusetshenziswa indlela yokubhala amanothi ye-ABC, imodeli izingxenye zengoma, ukuthuthuka komculo nokushintsha kwamandla, bese ikhipha umsebenzi ophelele ngesakhiwo.

Vakashela iwebhusayithi esemthethweni: vakashela ikhasi eliyisiqalo le-StepAudio 3 ku-https://static.stepfun.com/blog/stepaudio3/ ukuze ufunde ngamakhono emodeli ngayinye.
Khetha imodeli: khetha i-Realtime, ASR, TTS, Gen noma i-Music ngokuya ngezidingo.
Ngena esikhungweni sokuzama: vakashela isikhungo sokuzama sezwi se-Jieyue ku-https://www.stepfun.com/studio/audio ukuze uzame imiphumela yamamodeli ku-inthanethi.
Bhalisa epulatifomu evulekile: iya epulatifomu evulekile ye-Jieyue Xingchen ku-https://platform.stepfun.com/ ukuze ubhalise i-akhawunti futhi uthole i-API Key.
Bheka imibhalo ye-API: ekhasini lomhlahlandlela wokuxhuma wemodeli ehambisanayo epulatifomu evulekile, bheka amapharamitha ezicelo nezibonelo zokushaya ucingo.
Hlanganisa futhi ushayele: landela imibhalo ukuze uhlanganise i-API kuhlelo lwakho lokusebenza, bese udlulisa umbhalo, umsindo noma incazelo yolimi lwemvelo ukuze ukhiqize umphumela.
Lungisa futhi uthuthukise: lungisa umbala wezwi, imizwa, isivinini sokukhuluma, isakhiwo namanye amapharamitha ngokwemiphumela ebuyisiwe kuze kuhlangatshezwane nezidingo.

Imiphumela ezilinganisweni iqinile: ukunemba kokucabanga ngezwi kwe-Realtime (99.7%) kanye nezinga lamaphutha amagama le-ASR (1.7%) zombili zithole indawo yokuqala emhlabeni ku-Artificial Analysis, ziqinisekiswa inkampani ezimele.
Ukuxhumana okufana nokomuntu: iyakwazi ukubona ithoni, imizwa nejubane lengxoxo, yazi ukuthi kufanele iphendule nini, ilinde nini nokuthi ibhekane kanjani nokuphazanyiswa.
Iyakwazi ukukhuluma futhi yenze imisebenzi: ukucabanga, ukushaya amathuluzi nemisebenzi emide kungasebenza ngemuva ngokungavumelani, kungavimbi ingxoxo yamanje, kuguqule umsizi wezwi usuke ethuluzini lokuxoxa ube umsizi okwazi ukwenza umsebenzi.
Imodeli eyodwa iphethe lonke uchungechunge lokukhiqiza umsindo: ukudabula amazwi, imisindo, umsindo wemvelo nomculo wangemuva kwakudinga amathuluzi amaningi, kodwa manje ukukhiqiza okukodwa kunikeza zonke izingqimba zomsindo futhi kuhlele ngokunembile isikhathi nokulandelana kokuvela kwemisindo ehlukahlukene.
Umsebenzisi uyaphinde anikezwe amandla okudala: isekela izindlela eziningi zokufaka, ezihlanganisa ukucula ngaphandle komculo, amazwi engoma, umculo nendlela yokubhala amanothi ye-ABC; i-AI ikusiza ukuqhubeka nokuphelelisa umsebenzi.

Funda ngomkhiqizo: vakashela ikhasi eliyisiqalo le-StepAudio 3 ku-https://static.stepfun.com/blog/stepaudio3/, ujwayelane namakhono amamodeli amahlanu e-Realtime, ASR, TTS, Gen ne-Music.
Zame ku-inthanethi: ngena esikhungweni sokuzama sezwi se-Jieyue, faka umbhalo noma ulayishe umsindo, uzwe mahhala imiphumela yamamodeli.
Bhalisa epulatifomu: bhalisa i-akhawunti epulatifomu evulekile ye-Jieyue Xingchen, dala uhlelo lokusebenza futhi uthole i-API Key.
Bheka imibhalo: vula ikhasi lomhlahlandlela wokuxhuma wemodeli ehambisanayo ukuze ufunde ngekheli lesixhumi esibonakalayo, amapharamitha ezicelo nendlela yokukhokhisa.
Qalisa ukushaya ucingo: dlulisa umbhalo, ifayela lomsindo noma incazelo yolimi lwemvelo ngesicelo se-API ukuze uthole umphumela wokuhlanganiswa, ukuqaphela noma ukukhiqizwa kwenkulumo.
Lungisa amapharamitha: lungisa umbala wezwi, imizwa, isivinini sokukhuluma, ulimi, ukuhlelwa komsindo namanye amapharamitha ngokwemiphumela.
Hlanganisa futhi wethule: faka i-API emkhiqizweni noma ekuhambeni komsebenzi wakho, uqedele ukwethulwa futhi uqhubeke uqaphe imiphumela.

Iwebhusayithi yephrojekthi: https://static.stepfun.com/blog/stepaudio3/

Ukunakekelwa kwamakhasimende okuhlakaniphile nabasizi bezwi: ingxoxo yesikhathi sangempela yezindlela zombili, ekwazi ukubona imizwa yomsebenzisi nokwenza imisebenzi ngokungavumelani, okwenza ukuxhumana kwamakhasimende kube kwemvelo futhi kusebenze kahle.
Amaminithi emihlangano nokukhiqizwa kwemibhalo engezansi: i-ASR ibhala ngokunembile umsindo omude nezingxoxo zabantu abaningi, ibone ngokuzenzakalelayo amagama obuchwepheshe, bese ikhiqiza ngokuqondile amaminithi omhlangano noma imibhalo engezansi yevidiyo.
Ukuxhumana kwezwi emotweni: isekela izindawo ezinomsindo, iziphimiso zezilimi zesifunda kanye nenkulumo exubile yesiShayina nesiNgisi, iphendule ngokubambezeleka okuphansi futhi ithuthukise ulwazi lokulawula ngezwi ngesikhathi sokushayela.
Ukukhiqizwa kwamafilimu nokuqukethwe okulalelwayo: emidlalweni yomsakazo, ezincwadini ezilalelwayo nasekudabuleni amazwi opopayi, ikhiqiza ngesikhathi esisodwa izwi lomuntu, imisindo, umsindo wemvelo nomculo wangemuva, iqede isidingo sokubambisana kwamathuluzi amaningi nokuxuba umsindo kamuva.
Ukudala umculo nokukhiqiza izingoma eziculwa kabusha: ngokucula ngaphandle komculo, ihlela ngokuzenzakalelayo ukuhambisana komculo, noma iqhubeke nokubhala ingoma ephelele ngokusekelwe emazwini nasemculweni, isize abaculi baqede ngokushesha i-Demo nomsebenzi wokugcina.