Imibuzo Evame Ukubuzwa

Ukufaniswa kwezwi le-AI kusebenzisa ama-algorithm okufunda komshini ukuze kuhlaziywe izici ezihlukile zezwi lakho (i-timbre, iphimbo, i-accent, amaphethini okukhuluma). Ngemva kokucubungula inethiwekhi ye-neural, ikhiqiza inkulumo entsha enokufana okuphezulu kakhulu kwe-timbre, igcina izici zezwi lokuqala ngisho noma sikhuluma amagama ahluke ngokuphelele.

Ukuklona izwi nge-AI kwenzeka ngezinyathelo ezine eziyinhloko: Isinyathelo 1 - Ukuqoqa izwi, ulayisha amasampula omsindo acacile emizuzwana engu-3 kuya kwengu-15 njengezigxivizo zezwi; Isinyathelo 2 - Ukukhipha izici, ama-algorithm okufunda ngomshini ahlaziya ngokujulile izici ezihlukile zezwi lakho, okuhlanganisa umbala wezwi, ukuphakama kwezwi, imvamisa, ukwehla nokwenyuka kwezwi, isivinini sokukhuluma, i-accent, izindlela zokuphinyisa nesitayela sokukhuluma, akhe ama-vector ezici zezwi; Isinyathelo 3 - Ukuqeqesha imodeli, kusetshenziswa ukufunda okujulile nobuchwepheshe bamanethiwekhi e-neural ukuqeqesha imodeli, ifunde futhi ikhumbule amaphethini ezici zezwi lakho, isungule ubudlelwano bokumepha izwi; Isinyathelo 4 - Ukukhiqiza izwi, ukuguqula umbhalo ube inkulumo ngemodeli eqeqeshiwe, kukhiqizwe inkulumo entsha efana kakhulu nezwi lakho langempela, kugcinwe izici zezwi langempela ngisho nalapho kushiwo amagama ahluke ngokuphelele. Yonke inqubo isebenzisa ama-algorithm aphambili okufunda ngomshini nobuchwepheshe bamanethiwekhi e-neural, kuhlanganiswe nokuqeqeshwa ngedatha enkulu yezwi, ukuqinisekisa ukufana okuphezulu kakhulu kokuklona umbala wezwi nokubuyisela ngokunembile umbala wezwi, ukuphakama kwezwi nemininingwane yemizwa. Ubuchwepheshe bokuklona ngokushesha be-kikivoice bulungisa inqubo yakudala eyayidinga amahora okuqeqesha, buqede ukuklona izwi kwekhwalithi ephezulu ngaphakathi kwemizuzu emi-3 ngamasampula omsindo wemizuzwana embalwa kuphela.

Udinga umsindo ocacile, wekhwalithi ephezulu wemizuzwana engu-3 kuya kwengu-15 ukuze uklone izwi lakho ngempumelelo. Sincoma ukusebenzisa umsindo wemizuzwana engu-10 kuya kwengu-15 ukuze uthole ikhwalithi yokuklona engcono kakhulu. Uma ifayela lakho lomsindo lide, i-kikivoice inikeza usizo lokusika oluhlakaniphile: kumasampula angaphezu kwemizuzwana engu-20, uhlelo luzokhetha ngokuzenzakalelayo ingxenye engcono kakhulu yokuklona izwi; ungakwazi futhi ukusika mathupha ngokuhudula izibambo ukuze ukhethe izingxenye zenkulumo ecacile zemizuzwana engu-3 kuya kwengu-15, ngendlela elula necacile.

Izindlela ezingcono kakhulu: inkulumo ecacile engenamsindo ongemuva noma umculo; indlela yokukhuluma nesigqi semvelo; ukuqoshwa endaweni ethule; sebenzisa imakrofoni yekhwalithi enhle; imizuzwana engu-3 kuya kwengu-15 yenkulumo eqhubekayo. Uma umsindo wakho wokuqala mude, sincoma ukusebenzisa isici sosizo lokusika ukuze ukhethe ingxenye ecace kakhulu neyemvelo yemizuzwana engu-3 kuya kwengu-15, futhi sincoma imizuzwana engu-10 kuya kwengu-15 ukuze uthole imiphumela engcono kakhulu yokuklona.

kikivoice isekela amafomethi amaningi omsindo, okuhlanganisa WAV, MP3, M4A, AAC, OGG, OPUS, FLAC, WMA, ALAC, AIFF, AMR, nokunye, kanye namafomethi evidiyo MP4, MOV, MKV, AVI, WEBM, nokunye, ngomkhawulo omkhulu kasayizi wefayela ongu-50MB. Ukukhetha ifomethi kuvumelana nezimo. Okubaluleke kakhulu ukuqinisekisa ukuthi umsindo olayishiwe awunamsindo ongemuva, umsindo uhlanzekile, futhi inkulumo ifundwa ngokucacile nangokwemvelo ukuze kutholakale imiphumela engcono kakhulu yokuklona.

Yebo, i-kikivoice inomsebenzi wokurekhoda owakhelwe ngaphakathi osuselwe kusiphequluli. Uma uchofoza inkinobho yokuqala ukurekhoda, isiphequluli sizoveza isicelo sokugunyaza imvume yokurekhoda. Sicela uchofoze uqinisekise ukugunyaza ngaphambi kokurekhoda. Uma ukugunyaza kunqatshiwe, umsebenzi wokurekhoda ungase ungasebenzi kahle, kubangele ukuhluleka kokuqopha. Ngemva kokugunyazwa, khuluma endaweni ethule ukuze ubambe izwi lakho khona manjalo.

Inqubo yokuklona ihlukaniswe izinyathelo ezintathu: Isinyathelo 1 - layisha bese ukhetha umsindo wemizuzwana engu-3 kuya kwengu-15; Isinyathelo 2 - hlela okuqukethwe bese ukhetha imodeli; Isinyathelo 3 - qala umsebenzi wokuklona. Ngokuvamile iqedwa ngaphakathi kwemizuzu emi-3, isikhathi esiqondile sincike ebudeni bokuqukethwe, imodeli yokuklona ekhethiwe, nomthwalo wokucubungula wamaseva e-AI.

Ukuzama mahhala akudingi ukubhalisa i-akhawunti, futhi akudingeki ubophe ikhadi lesikweletu. Ungalayisha umsindo ngqo uqale ukuklona ngokushesha, uzame izici eziyinhloko ngaphandle kokubhalisa noma ukungena ngemvume. Idatha yakho yomsindo ayigcinwa unomphela, isuswa ngokuzenzakalelayo ngemva kokucubungulwa, futhi ungayisusa mathupha ngemva kokuba umsebenzi usuqediwe, okuqinisekisa ubumfihlo nokuvikeleka. Isikhathi sokuguqula sifushane, ngokuvamile siqedwa ngaphakathi kwemizuzu emi-3. Umsindo okhiqiziwe ungalandwa izikhathi ezingenamkhawulo futhi ungawulanda noma nini. Uma izici zokungena ngemvume/zokubhalisa zitholakala, zenzelwe ukuphatha kalula idatha nezilungiselelo zephrojekthi yakho yokuklona, okuhlanganisa ukubuka umlando, ukugcina izilungiselelo ezivamile, ukuvumelanisa ukusetshenziswa kwamadivayisi amaningi, nokunye.

Uma nje ukufinyelela ngesiphequluli kusekelwa, ungayisebenzisa. Akudingeki ulande uhlelo lokusebenza. i-kikivoice isekela wonke amadivayisi anesiphequluli sesimanje, okuhlanganisa i-Windows, Mac, iOS, Android, nokunye. Noma usebenzisa ideskithophu, ilaptop, ithebulethi noma ifoni, uma unesiphequluli ungasebenzisa isevisi yethu yokuklona izwi.

Yebo, sinikeza izinga lamahhala lokuzama izici eziyinhloko. Amaphuzu ekhredithi okuzama mahhala asethwa kabusha masonto onke, futhi amakhredithi asetshenziswa ngesikhathi sokuguqula. Ungayisebenzisa ngokuzethemba. Idatha yezwi lakho ibethelwa futhi isuswa ngokuzenzakalelayo ngemva kokucubungulwa, kusekelwa kokubili ukususa okuzenzakalelayo nokususa mathupha, okuqinisekisa ubumfihlo nokuvikeleka. Sinikeza amamodeli amaningi akhelwe ngaphakathi okuklona izwi: Kiki Core (ilula futhi izinzile, ifanele izidingo ezisheshayo zansuku zonke), Kiki Pro (inemizwa ecebile, inezinketho eziningi zezilungiselelo, ifanele ukudala okuqukethwe kwezinga lochwepheshe), Kiki Multilingual (isekela izilimi ezingaphezu kwezingu-75, imiphumela izinzile, ifanele ukukhiqiza okuqukethwe komhlaba wonke). Ungakhetha ngokuguquguquka ngokwezidingo ezahlukene.

i-kikivoice iyiwebhusayithi yokufaniswa kwezwi le-AI yabadali abangochwepheshe, egxile ekufaniseni izwi le-AI okusheshayo nokusebenzayo. Sinamamodeli angu-3 akhelwe ngaphakathi wokufaniswa kwezwi le-AI ukuze sivumelane nezidingo ezahlukene zenqubo yomsebenzi: Imodeli ye-Kiki Core ilinganiselwe futhi izinzile, isivinini sokukhiqiza esisheshayo, izwi elingokoqobo, isekela izilimi ezingu-10+, ilungele ukwakhiwa kokuqukethwe kwazo zonke izinhloso nezidingo ezisheshayo zansuku zonke; Imodeli ye-Kiki Pro isezingeni lobuchwepheshe, izwi elingokoqobo kakhulu, isekela izilimi ezingu-8+, inikeza ukulawula kwemizwa okungu-15+, ilungele imisebenzi yezinga le-studio nokwakhiwa kokuqukethwe kwezinga lobuchwepheshe; Imodeli ye-Kiki Multilingual isekela izilimi ezingu-75+, isivinini sokukhiqiza esisheshayo, ilungele okuqukethwe kokwenziwa kwasekhaya komhlaba wonke namaphrojekthi wezilimi eziningi. Imiphumela yekhwalithi ephezulu ingatholakala ngemizuzwana embalwa nje yomsindo, ngokuvamile iqedela imisebenzi yokufanisa phakathi kwemizuzu emi-3. Sisebenzisa ama-algorithm okufunda komshini athuthukile nobuchwepheshe benethiwekhi ye-neural ukuze siqinisekise ukufana okuphezulu kwe-timbre nokubuyisela ngokunembile i-timbre, iphimbo, nemininingwane yemizwa.

Izinga lamahhala lisekela ibanga lezinhlamvu ezingu-500-2000 ekuguqulweni ngakunye. Amamodeli ahlukene okuklona asekela ubude bombhalo obuphezulu obuhlukile: amamodeli e-Kiki Core ne-Kiki Multilingual ngokuvamile asekela umbhalo omude, kuyilapho imodeli ye-Kiki Pro inemikhawulo yobude bombhalo ehambisanayo ngesikhathi igcina ikhwalithi ephezulu yokukhipha. Imikhawulo eqondile izohluka kuye ngemodeli oyikhethayo. Izinhlelo ze-premium zinikeza imikhawulo yezinhlamvu ephezulu nokucubungula okubekwa phambili.

Ikhwalithi yomsindo ofakiwe ithinta ngqo ikhwalithi ephumayo. Ukuthi inkulumo icacile yini, ukuthi kukhona umsindo noma umsindo ongemuva yini, ubude besampula, noma ikhwalithi yemakrofoni, konke kuzothinta imiphumela yokuklona. Sincoma ukuzama futhi ngokuqoshwa okucace kakhudlwana, noma ukusebenzisa isici sosizo lokusika ukukhetha izingxenye zomsindo ezicace kakhudlwana, noma ukukhetha amamodeli ahlukene okuklona e-kikivoice (Kiki Core, Kiki Pro, Kiki Multilingual) ukuze uzame ukulungiswa kwemiphumela yemininingwane ehlukile. Sekukonke, ukufana kwethu kokuklona umbala wezwi kuphezulu kakhulu.

Qopha endaweni ethule, sebenzisa imakrofoni enhle, futhi unikeze umsindo ohlanzekile wemizuzwana engu-3 kuya kwengu-15. Lapho ufunda, okuqukethwe kufanele kucace, ukuphimisa kube okunembile, isivinini sokukhuluma sibe esimaphakathi, inkulumo ibe yemvelo, kugwenywe inkulumo engacacile noma eshesha/ehamba kancane kakhulu. Khetha izingxenye zomsindo ezicacile nezingenamsindo ukuze uqinisekise ukuthi igama ngalinye liyazwakala, ukuze i-AI ifunde kangcono izici zezwi lakho futhi ikhiqize amazwi akloniwe ekhwalithi ephezulu.

Umsindo womshini ngokuvamile uvela ezinkingeni zekhwalithi yomsindo ofakiwe, njengenkulumo engacacile, umsindo, ukuqoshwa okunomsindo, amasampula amafushane kakhulu, noma ikhwalithi embi yemakrofoni. Sincoma ukuzama futhi ngokuqoshwa okucace kakhudlwana, noma ukusebenzisa isici sosizo lokusika ukukhetha izingxenye zomsindo ezicace kakhudlwana, noma ukukhetha amamodeli ahlukene okuklona e-kikivoice ukuze uzame ukulungiswa kwemiphumela yemininingwane ehlukile. Sekukonke, ukufana kwethu kokuklona umbala wezwi kuphezulu kakhulu.

i-kikivoice inokufana okuphezulu kokufaniswa kwe-timbre, ikhiqiza amakhophi angokoqobo acishe angahlukaniseki nezwi lokuqala, ibuyisela ngokunembile izici ze-timbre. Amamodeli ahlukene anemiphumela eyahlukene: Imodeli ye-Kiki Core ilinganiselwe futhi izinzile, isivinini sokukhiqiza esisheshayo, izwi elingokoqobo, isekela izilimi ezingu-10+, ilungele ukwakhiwa kokuqukethwe kwazo zonke izinhloso nezidingo ezisheshayo zansuku zonke; Imodeli ye-Kiki Pro isezingeni lobuchwepheshe, izwi elingokoqobo kakhulu, isekela izilimi ezingu-8+, inikeza ukulawula kwemizwa okungu-15+, ilungele imisebenzi yezinga le-studio nokwakhiwa kokuqukethwe kwezinga lobuchwepheshe; Imodeli ye-Kiki Multilingual isekela izilimi ezingu-75+, isivinini sokukhiqiza esisheshayo, ilungele okuqukethwe kokwenziwa kwasekhaya komhlaba wonke namaphrojekthi omhlaba wonke.

Yebo, ungakhuphula noma ube nenhle isivinini sokukhuluma ngezilungiselelo zokulawula izwi.

Yebo, i-kikivoice inganquma imizwa ngokuzenzakalelayo ngokusekelwe kokuqukethwe okufakiwe futhi ibambe ubuqiniso bemizwa. Amamodeli ahlukene anokusekelwa kwemizwa okuhlukile: imodeli ye-Kiki Pro inikeza izilawuli zemizwa ezingaphezu kwezingu-15 nezilawuli zokuqina kwemizwa, okukuvumela ulungise kahle amandla okuveza imizwa; amamodeli e-Kiki Core ne-Kiki Multilingual nawo asekela ukuvezwa kwemizwa okuyisisekelo. Ungalawula ukuvezwa kwemizwa nokuqina kwemizwa ngezilayida ezenziwe ngokwezifiso.

Imiphumela yokuklona izwi inqunywa kakhulu ikhwalithi yokuqukethwe okufakiwe. Ungalungisa imininingwane yemizwa nezici zezwi ngokukhetha amamodeli ahlukene okuklona: imodeli ye-Kiki Pro inikeza ukulawulwa kwemizwa okunembile kakhudlwana nezinketho zokulungisa ukuphakama kwezwi, ifanele amaphrojekthi adinga ukulawula okunembile; amamodeli e-Kiki Core ne-Kiki Multilingual nawo asekela ukulawulwa okuyisisekelo kokuphakama kwezwi. Ukulawula ukuphakama kwezwi kukuvumela uphakamise noma wehlise imvamisa yezwi ukuze ihambisane nezidingo zakho zokudala.

Yebo — sisekela izilimi ezingu-75+. kikivoice ihlanganisa amamodeli amathathu ezimo ezahlukene zokusetshenziswa: Kiki Core yokuqukethwe kwansuku zonke okusheshayo nokulinganiselayo; Kiki Pro yekhwalithi yezinga lesitudiyo enokufana kwezwi okungcono; kanye ne-Kiki Multilingual yokwenziwa kwasendaweni emhlabeni jikelele ngokuhlanganisa izilimi okubanzi kakhulu (75+). Sisekela izilimi ezibalulekile njengesiNgisi, iSpanishi, isiShayina, isiHindi, isiBengali, isiFulentshi, isiJalimane, isiJapane, isiKorea, isiPutukezi, isiNtaliyane, isi-Arabhu, nesi-Urdu — nezinye eziningi ngaphandle kwalokho. Khetha imodeli efana nekhwalithi yephrojekthi yakho, isivinini, nezidingo zokwenziwa kwasendaweni.

Yebo. Imodeli ye-Kiki Multilingual isekela kuze kufike ezilimini ezingaphezu kwezingu-75, ivumele izwi lakho lesiNgisi livumelane nezilimi eziningi ngaphandle kokulahlekelwa umbala wezwi wokuqala. Amamodeli ahlukene asekela izinombolo ezahlukene zezilimi, kodwa izilimi ezisetshenziswa kakhulu zisekelwa ngokuyisisekelo.

Klona izwi lakho ngolimi olulodwa, bese i-AI ilivumelanisa ngokuzenzakalelayo nezilimi ezisekelwayo kuyilapho igcina ukufana kombala wezwi nesigqi sokuphakama kwezwi. Kusekelwa ubuningi bezilimi ezingaphezu kwezingu-75. Amamodeli ahlukene asekela izinombolo ezahlukene zezilimi, kodwa izilimi ezisetshenziswa kakhulu zisekelwa ngokuyisisekelo.

Okwamanje, amathegi e-SSML awasekelwa. Ungasebenzisa isici somhleli sokufaka ikhefu ukuze ungeze amakhefu enziwe ngokwezifiso. I-AI izocubungula imizwa nezinkulumo ngokwemvelo ngokusekelwe kokuqukethwe kombhalo.

Umhleli usekela umsebenzi wokufaka ikhefu. Chofoza inkinobho yokufaka ikhefu ukuze ufake ithegi yekhefu endaweni yekhesa. Ungakhetha amakhefu ajwayelekile (ikhefu elifushane lamasekhondi angu-0.5, ikhefu elijwayelekile lesekhondi elingu-1.0, ikhefu elide lamasekhondi angu-3.0), noma wenze ngokwezifiso ubude bekhefu elingu-0 kuya ku-10 amasekhondi ngesilayida; ifomethi ithi ((=milliseconds)), isibonelo ((=1000)) kusho ikhefu lesekhondi elingu-1. I-AI izocubungula imizwa nezinkulumo ngokwemvelo ngokusekelwe kokuqukethwe kombhalo, ikhiqize isigqi senkulumo semvelo.

kikivoice inikeza i-Kiki Multilingual (izinga lomhlaba wonke), i-Kiki Core (izinzile futhi iyashesha), ne-Kiki Pro (ikhwalithi yezinga le-studio) ukuze ihlanganise izimo ezahlukene.

Khetha ngokwezidingo zephrojekthi: imodeli ye-Kiki Core ilinganiselwe futhi izinzile, isivinini sokukhiqiza siyashesha, izwi lingokoqobo, isekela izilimi ezingaphezu kwezingu-10, ifanele ukudala okuqukethwe okujwayelekile nezidingo ezisheshayo zansuku zonke; imodeli ye-Kiki Pro isezingeni lochwepheshe, izwi lingokoqobo kakhulu, isekela izilimi ezingaphezu kwezingu-8, inikeza izilawuli zemizwa ezingaphezu kwezingu-15, ifanele imisebenzi yezinga le-studio namaphrojekthi ezinga lochwepheshe; imodeli ye-Kiki Multilingual isekela izilimi ezingaphezu kwezingu-75, isivinini sokukhiqiza siyashesha, ifanele okuqukethwe kokwenza kube sendaweni yomhlaba wonke namaphrojekthi ezilimi eziningi. Indlela amakhredithi asebenza ngayo: Isinyathelo 1 - Qala umsebenzi omusha, chofoza ukuze uqale umsebenzi omusha wokuklona izwi; Isinyathelo 2 - Ukuhlola okuzenzakalelayo, uhlelo luqinisekisa ukuthi unamakhredithi anele; Isinyathelo 3 - Ukukhiqiza izwi, amakhredithi ashintshaniswa ngamandla ekhompiyutha e-AI. Indlela yokubala amakhredithi: Umsebenzi ngamunye wokuklona izwi nge-AI udinga amakhredithi, amakhredithi adonswa ngokwenani lezinhlamvu embhalweni ofakiwe, futhi imodeli ngayinye inomphindaphindi osuselwe emandleni ekhompiyutha adingekayo. Abaphindaphindi bamakhredithi bamamodeli ahlukene: Kiki Core ingu-2x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-200), Kiki Pro ingu-3x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-300), imodi esheshayo ye-Kiki Multilingual ingu-1x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-100), imodi yekhwalithi ephezulu ingu-2x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-200).

Imodeli ye-Kiki Core ilinganiselwe futhi izinzile, isivinini sokukhiqiza esisheshayo, izwi elingokoqobo, isekela izilimi ezingu-10+, isiphindaphindiseli samaphuzu singu-2x. Isivinini sokukhiqiza esisheshayo sisho ukubambezeleka okuphansi, okulungele ama-demo asheshayo, ukuphindaphinda, nokwakhiwa kokuqukethwe kwazo zonke izinhloso.

Kusekelwa ubuningi bezilimi ezingaphezu kwezingu-75. Amamodeli ahlukene asekela izinombolo ezahlukene zezilimi: imodeli ye-Kiki Multilingual isekela izilimi eziningi kakhulu (ezingaphezu kwezingu-75), imodeli ye-Kiki Core isekela izilimi ezingaphezu kwezingu-10, futhi ukusekelwa kwezilimi kwemodeli ye-Kiki Pro kuyahlukahluka ngenguqulo. Izilimi ezisetshenziswa kakhulu (njengesiNgisi, isiShayina, isiSpanishi, isiFulentshi, isiJalimane, isiJapane, isiKorea, isiNtaliyane, isiPutukezi, isiRashiya, njll.) zisekelwa ngokuyisisekelo. Izwi lakho lingavumelaniswa nanoma yiluphi ulimi olusekelwayo.

Yebo, ungakhetha imodeli ehlukile kuseshini ngayinye yokufanisa kuye ngezinhloso zakho.

Imodeli ye-Kiki Pro isezingeni lobuchwepheshe, izwi elingokoqobo kakhulu, isekela izilimi ezingu-8+, inikeza ukulawula kwemizwa okungu-15+, isivinini sokukhiqiza esiphakathi, isiphindaphindiseli samaphuzu singu-3x. Ilungele kakhulu imisebenzi yezinga le-studio, njenge-voiceover yezentengiso, ama-audiobook wekhwalithi ephezulu, ukukhiqizwa kwemithombo yezindaba yobuchwepheshe nezinye amaphrojekthi adinga inkulumo yemizwa ecacile nokukhishwa kwekhwalithi ephezulu.

Izimo zokusetshenziswa zibanzi: ukudubhula ividiyo, ukukhiqiza ama-podcast, imfundo eku-inthanethi, izincwadi ezilalelwayo, ukulingisa amazwi abalingiswa bemidlalo, ama-voiceover ezikhangiso, okuqukethwe kwezokuxhumana, ukuqeqeshwa kwezinkampani, ukusakaza izindaba, ukuxoxa amadokhumentari, ukudubhula ama-animation, abasakazi ababonakalayo, ukudala okuqukethwe ngezilimi eziningi, izingcingo zesevisi yamakhasimende, abasizi bezwi, ukuhamba ngezwi, nokunye. Ifanele noma iyiphi iphrojekthi edinga ukukhiqiza ngokushesha amazwi abantu ekhwalithi ephezulu.

Yebo, ifanelekile ukulandisa kwevidiyo, okuqukethwe kwenkundla yezokuxhumana, nokukhiqizwa kwemithombo yezindaba yobuchwepheshe.

Yebo, i-Multilingual ivumela izwi lakho likhulume noma yiluphi ulimi ngokugeleza, iphule izithiyo zolimi kubabukeli bomhlaba wonke.

Yebo, ungakwazi ukulawula ukukhethwa kwemodeli, ukufakwa kombhalo, ulimi, isivinini sokukhuluma, ukuphakama kwezwi nemizwa kuphaneli eyodwa yokulawula.

Yebo. Umhleli usekela umsebenzi wokufaka ikhefu. Chofoza inkinobho yokufaka ikhefu ukuze ufake ithegi yekhefu endaweni yekhesa. Ungakhetha amakhefu ajwayelekile (ikhefu elifushane lamasekhondi angu-0.5, ikhefu elijwayelekile lesekhondi elingu-1.0, ikhefu elide lamasekhondi angu-3.0), noma wenze ngokwezifiso ubude bekhefu elingu-0 kuya ku-10 amasekhondi ngesilayida. I-AI izocubungula imizwa nezinkulumo ngokwemvelo ngokusekelwe kokuqukethwe kombhalo, ikhiqize isigqi senkulumo semvelo. Lezi zici zikuvumela ulungise kahle amakhefu nesigqi kumsindo.

Yebo, kusekelwa ukulanda okungenamkhawulo nokudlala okungenamkhawulo. Amafomethi okuthekelisa nokulanda asekelwa ezinhlobo ezi-5: MP3, WAV, OGG, AAC, OPUS. Ikhwalithi yomsindo ingakhethwa njengejwayelekile noma ikhwalithi ephezulu, kube lula ukuyisebenzisa kumaphrojekthi.

Lindela okukhiphayo kwezinga lochwepheshe, ukucubungula okusheshayo, nezinketho eziningi zamafomethi ukuze kufezeke imiphumela yezinga lezohwebo.

Yebo, ungazama izilimi, izitayela, noma amamodeli ahlukene ukuze wenze ushintsho olungenamkhawulo.

Ngaphakathi kwamaphuzu ekhredithi okuzama mahhala asethwa kabusha masonto onke, ungakhiqiza okuqukethwe njengoba kudingeka, futhi ukuguqulwa ngakunye kuzodla amaphuzu ekhredithi ahambisanayo.

Okwamanje, asinikezi isixhumi esibonakalayo se-API. Sihlela ukwethula izinsizakalo ze-API esikhathini esizayo, sivumele abathuthukisi bahlanganise ngqo amakhono okuklona e-kikivoice ezinhlelweni zabo zokusebenza, okuhlanganisa ukusekelwa kokucubungula ngobuningi, ukumiswa kwamapharamitha ngokwezifiso, nezinye izici.

Okwamanje, sisekela ukufaka umbhalo. Kumaskripthi amade, sincoma ukuwahlukanisa abe ngobude bezinhlamvu obufanele (ngokuvamile ngaphakathi kwebanga lezinhlamvu ezingu-500-2000) ngokusekelwe kwimodeli yokuklona oyikhethayo.

Idatha yezwi ibethelwa futhi isuswa ngokuzenzakalelayo ngemva kokuphela kweseshini. Umsindo olayishiwe ungasuswa mathupha ngokuchofoza ukususa kusixhumi sewebhu sokuklona kwe-AI, kusekelwa kokubili ukususa okuzenzakalelayo nokususa mathupha.

Kuphephile, sisebenzisa ukubethela nezinyathelo zokuphepha ezinezendlalelo ukuvikela ubumfihlo bokulayisha kwakho. Umsindo olayishiwe ungasuswa mathupha ngokuchofoza ukususa kusixhumi sewebhu sokuklona kwe-AI, kusekelwa kokubili ukususa okuzenzakalelayo nokususa mathupha.

Cha, sisebenzisa idatha yakho kuphela ukudala ukufaniswa futhi ngeke siyisebenzise ukuqeqesha amanye amamodeli.

Cha, ukufaniswa kudinga amasampula womsebenzisi ngokwakhe, futhi sinokuvikelwa ukuvimbela ukufaniswa okungagunyaziwe.

Cha, imigomo yethu yesevisi ivimbela ngokuqinile ukuklona amazwi abanye ngaphandle kwemvume. Ukuze uthole imithethonqubo enemininingwane, sicela ubuke imigomo yethu yesevisi nenqubomgomo yobumfihlo ukuze uthole imininingwane eyengeziwe.

Kuphela ngemvume yabo ecacile - kufanele ube nemvume yokusebenzisa izwi lomunye umuntu. Ukuze uthole imithethonqubo enemininingwane, sicela ubuke imigomo yethu yesevisi nenqubomgomo yobumfihlo ukuze uthole imininingwane eyengeziwe.

Idatha yakho isuswa ngokuzenzakalelayo ngemva kokucubungulwa, futhi asiyigcini unomphela. Umsindo olayishiwe ungasuswa mathupha ngokuchofoza ukususa kusixhumi sewebhu sokuklona kwe-AI, kusekelwa kokubili ukususa okuzenzakalelayo nokususa mathupha.

Yebo, sibethela ukulayishwa kwamazwi futhi silandela izivumelwano zobumfihlo eziqinile kuyo yonke inkundla. Umsindo olayishiwe ungasuswa mathupha ngokuchofoza ukususa kusixhumi sewebhu sokuklona kwe-AI, kusekelwa kokubili ukususa okuzenzakalelayo nokususa mathupha.

Sizibophezele ku-AI enesimilo futhi sivimbela ngokuqinile ukudalwa kokuqukethwe okunonya, okuhlambalazayo noma okukhohlisayo. Ukuze uthole izinqubomgomo ezinemininingwane, sicela ubuke imigomo yethu yesevisi nenqubomgomo yobumfihlo ukuze uthole imininingwane eyengeziwe.

Isinyathelo 1 - Layisha isampula yezwi: Layisha ifayela lomsindo elicacile lemizuzwana engu-3 kuya kwengu-15 noma urekhode ngqo, bese ukhetha umsindo njengesampula yezwi. Isinyathelo 2 - Yenza izilungiselelo ngokwezifiso: Faka okuqukethwe kombhalo, khetha izilimi ezingaphezu kwezingu-75, khetha imodeli yokuklona, lungisa isivinini sokukhuluma namapharamitha okuzinza. Isinyathelo 3 - Khiqiza bese ulanda: Chofoza ukukhiqiza ukuze uqale umsebenzi wokuklona, ngokuvamile uqedwa ngaphakathi kwemizuzu emi-3, kusekelwa ukudlala nokulanda okungenamkhawulo, amafomethi okuthekelisa asekelwa ezinhlobo ezi-5: MP3, WAV, OGG, AAC, OPUS, ikhwalithi yomsindo ingakhethwa njengejwayelekile noma ikhwalithi ephezulu.

Izinga lamahhala lifaka amaphuzu ekhredithi asethwa kabusha masonto onke, amakhredithi asetshenziswa ngesikhathi sokuguqula, nokufinyelela kuwo wonke amamodeli. Ngaphambi kokusebenzisa, sicela uqinisekise ukuthi unamalungelo okukopisha namalungelo okusebenzisa umsindo olayishiwe. Ukuze uthole imininingwane eningiliziwe, sicela ubheke isivumelwano sethu somsebenzisi nenqubomgomo yobumfihlo.

Izinhlelo ze-premium zinikeza imikhawulo yezinhlamvu ephezulu nokucubungula okubekwa phambili. Isevisi yemodeli isekela ukusetshenziswa kwezohwebo, kodwa kudingeka uqinisekise ukuthi unamalungelo okusebenzisa umsindo olayishiwe kwezohwebo. Ukuze uthole imininingwane eningiliziwe, sicela ubheke isivumelwano sethu somsebenzisi nenqubomgomo yobumfihlo.

Isevisi yemodeli ye-kikivoice isekela ukusetshenziswa kwezohwebo. Ngaphambi kokusebenzisa, sicela uqinisekise ukuthi unamalungelo okukopisha namalungelo okusebenzisa umsindo olayishiwe. Amazwi akloniwe akhiqiziwe angasetshenziswa kumaphrojekthi ezohwebo, inqobo nje uma umsindo wokuqala owulayishile wona unamalungelo okusetshenziswa kwezohwebo. Ukuze uthole imininingwane eningiliziwe, sicela ubheke isivumelwano sethu somsebenzisi nenqubomgomo yobumfihlo.

Okwamanje, abasebenzisi bangakwazi ikakhulukazi ukuzama izici eziyinhloko mahhala ngaphandle kokungena ngemvume. Amaphuzu ekhredithi okuzama mahhala asethwa kabusha masonto onke akuvumela uqale ukusebenzisa ngokushesha. Esikhathini esizayo, sihlela ukuhlinzeka ngezinhlelo zokubhalisa zamazinga ahlukene, okuhlanganisa izinhlelo zokubhalisa zanyanga zonke, ukuze kuhlangatshezwane nezidingo ezahlukene zabasebenzisi. Kubasebenzisi abanesilinganiso esikhulu, sihlela ukunikeza izinketho ezenziwe ngokwezifiso ngezinhlelo zebhizinisi esikhathini esizayo. Ukuze uthole imininingwane eningiliziwe, sicela ubheke isivumelwano sethu somsebenzisi nenqubomgomo yobumfihlo.

Ukusetshenziswa kubalwa ngokwenani lezinhlamvu embhalweni owufakayo ukuze uwukhiqize. Umsebenzi ngamunye wokuklona izwi nge-AI udinga amakhredithi, futhi amakhredithi adonswa ngokwenani lezinhlamvu embhalweni ofakiwe. Imodeli ngayinye inomphindaphindi osuselwe emandleni ekhompiyutha adingekayo: Kiki Core ingu-2x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-200), Kiki Pro ingu-3x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-300), imodi esheshayo ye-Kiki Multilingual ingu-1x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-100), imodi yekhwalithi ephezulu ingu-2x (isb., umsebenzi wezinhlamvu ezingu-100 udla amakhredithi angu-200).

Yebo, ungakhansela ukubhalisa kwakho noma nini futhi ugcine ukufinyelela kuze kube sekupheleni komjikelezo wamanje wokukhokha. Ukuze uthole imininingwane eningiliziwe, sicela ubheke isivumelwano sethu somsebenzisi nenqubomgomo yobumfihlo.

Yebo, i-kikivoice ikuvumela ukuthi udale ama-voiceover amaphrofeshinali ngaphandle kwezikhathi ze-studio ezibizayo.

Nakanjani - khiqiza amazwi abalingiswa ahlukile emidlalweni, ama-mod, nokuhlangenwe nakho okusebenzisanayo.

Yebo, dala ukulandisa kwezifundo nezinto zokuqeqesha ngamazwi angaguquki, azwakalayo.

Yebo, abadali abaningi basebenzisa i-kikivoice ukukhiqiza ama-podcast ngokushesha nangokungaguquki.

Yebo, dala okuqukethwe kwe-TikTok, Instagram, YouTube nezinye izinkundla zokuxhumana ngezwi lakho elikloniwe.

Vuselela ikhasi, hlola ukuxhumana kwakho, layisha kabusha isampula, noma mfushane umbhalo. Xhumana nokwesekwa uma izinkinga ziqhubeka.

Hlola ivolumu yedivayisi, izimvume zesiphequluli, qinisekisa ukuthi ama-headphone axhunyiwe, bese ukhuphula ivolumu yethebhu.

Ukucubungula kuncike emthwalweni weseva, ikhwalithi yomsindo, nosayizi wefayela. Izikhathi ezimatasa zingangeza imizuzu embalwa eyengeziwe.

Imiphumela evalekile isho umsindo ongemuva noma ukuqoshwa kwekhwalithi ephansi. Qopha kabusha endaweni ethule usebenzisa imakrofoni engcono.

Nikeza amasampula acace kakhudlwana, zama umbhalo womthombo ohlukile, noma lungisa izilungiselelo zezwi ukuze ukuphimisa kube bushelelezi.

Sebenzisa igumbi elithule, nikeza imizuzwana engu-3 kuya kwengu-15 yenkulumo yemvelo, bese ulayisha kabusha usebenzisa imakrofoni yekhwalithi.

Susa amazwi afanisiwe ezilungiselelweni ze-akhawunti yakho noma uthembele ekususweni okuzenzakalelayo ngemva kokucubungulwa.

Sisebenzisa amanethiwekhi e-neural asezingeni eliphezulu, ukufunda komshini, nokuhlaziywa kwamaphethini wezwi.

Sihlaziya izici zezwi, siqeqesha amanethiwekhi e-neural phezu kwazo, bese sihlanganisa inkulumo efanisiwe.

Yebo, sisebenzisa i-AI esezingeni eliphezulu namanethiwekhi e-neural ukuklona amazwi ngokunemba okungu-99%.

Amanethiwekhi e-neural athuthukisiwe nama-algorithm aphambili akhiqiza amakloni ngemizuzwana, kwesinye isikhathi ngokubambezeleka okungaphansi kuka-50ms.

Ukufanisa ngaso leso sikhathi, okungokoqobo ngemizwa, ukusekelwa kwezilimi eziningi, ubumfihlo kuqala, izilawuli ezilula, futhi kumahhala.

i-kikivoice isesigabeni se-MVP futhi sifuna ukwenza ukufaniswa kwezwi okuthuthukile kufinyeleleke kuwo wonke muntu.

Cha, asinawo izindleko ezifihliwe. Akukho ukubhalisa, alikho ikhadi lesikweletu, azikho izithiyo ze-premium.

Abadali abangaphezu kuka-10,000 bathembela ku-kikivoice ngezidingo zabo zokufanisa.

Idatha yezwi ebethelwe, ukususwa okuzenzakalelayo, izinyathelo zokuvikela ezakhelwe ngaphakathi, futhi akukho ukwabelana nabanye.

Cha, i-kikivoice yakhelwe ukuba iqondakale kalula. Landela inqubo yezinyathelo ezi-3 ukuze uthole imiphumela ngokushesha.

Yebo, finyelela i-kikivoice kunoma yisiphi isiphequluli kuma-smartphone nakuma-tablet.

Noma yisiphi isiphequluli sanamuhla siyasebenza—i-Chrome, Firefox, Safari, ne-Edge zisekelwa ngokugcwele.

Cha, i-kikivoice isekelwe kuwebhu. Vele uvule isiphequluli bese uqala ukufanisa ngaso leso sikhathi.

Yebo, dala futhi uphathe amazwi afanisiwe amaningi ngokudingekayo ngaphandle kwemikhawulo.

Thintana [email protected] noma usebenzise ifomu lokuxhumana kuwebhusayithi.

Hlola i-FAQ, okokufundisa, noma ufinyelele kosekelo ukuze uthole isiqondiso esiningiliziwe.

Yebo, impendulo yamukelekile—vele usithumelele umlayezo ngefomu losekelo.

Impela, sihlala sikhipha izici ezintsha nokuthuthukiswa ngemuva kwesehlakalo.