Ukwenza umsindo wemvelo wezilimi ezimbili oseduze usebenzisa i-ElevenLabs, i-Whisper ne-ffmpeg

Uhlolojikelele lwephrojekthi

Le ndlela yokusebenza isetshenziselwa ukwenza umsindo osondelene wezilimi ezimbili, isiShayina nesiNgisi, ozizwakala ngokwemvelo futhi ugeleza kahle. Umgomo oyinhloko akukhona ukwenza imodeli yokuguqula umbhalo ibe yinkulumo ilingise yonke imisindo, kodwa ukuhlukanisa umphumela ube ngamathrekhi anemisebenzi ehlukene: i-ElevenLabs iphethe amazwi nokuphefumula, izinto zangempela ziphathe imisindo yokuqabula neyendawo, i-ffmpeg iphathe ukusika, ukushintsha ijubane, ukuxuba nokuhlanganisa, kanti i-Whisper yona ibe ithuluzi lokuhlola ikhwalithi.

Isiphetho esibaluleke kakhulu: Imodeli yezwi kufanele iphathe kuphela “amazwi” nokuphefumula okuhlanganiswe nesigqi sokukhuluma. Imisindo yokuqabula, imisindo emanzi neminye imisindo engelona ulimi kufanele yenziwe ngokuzimela, bese ixutshwa nezwi.

Inqubo eku-README yaqedwa ku-Mac mini, kodwa izindlela ezisetshenzisiwe azincikile ohlotsheni oluthile lwe-Mac. Uma nje uhlelo lukwazi ukubiza i-ElevenLabs API, ukusebenzisa i-ffmpeg nokusebenzisa i-Whisper ekubhaleni amazwi, izinyathelo eziyinhloko zingaphindwa.

Amathuluzi ayinhloko nemisebenzi yawo

• ElevenLabs: Yakha umbala wezwi nge-Voice Design, bese isebenzisa imodeli yokuguqula umbhalo ibe yinkulumo ukwenza amazwi.
• ffmpeg: Ihlanganisa izinga lokusampula, yehlisa ijubane lokukhuluma, isika izinto, yenza ukungena nokuphuma kancane, ixube futhi ihlanganise umsindo.
• Whisper: Ihlola ukuthi izinto ezingelona ulimi aziguqulwa yini yimodeli zibe amagama abonakalayo.
• Amarekhodi angempela anemvume esemthethweni: Ahlinzeka ngemisindo yendawo ezimele. Inqubo yokuqala isebenzisa kuphela izinto ze-CC0 ezisendaweni yomphakathi, futhi igcina imithombo yazo.

Ukufakwa nokulungiselela ngaphambi kokuqala

1. Lungiselela i-ElevenLabs

Bhalisa ku-ElevenLabs bese uthola imininingwane yokufinyelela ekwazi ukubiza i-Voice Design kanye ne-API yokuguqula umbhalo ibe yinkulumo. Ama-interface adingekayo empeleni afaka POST /v1/text-to-voice/design kanye POST /v1/text-to-voice.

Esinyathelweni sokuqala thumela voice_description nombhalo wokulalela, ukuze uthole izibonelo eziningi ezizokhethwa; ngemva kokukhetha esisodwa, sebenzisa i-generated_voice_id yaso ukugcina umbala wezwi osemthethweni. Ngemva kokugcina ngempumelelo, rekhoda i-voice_id; zonke izinqubo zokuhlanganisa ezilandelayo kufanele zisebenzise le ID.

2. Faka i-ffmpeg

Qinisekisa ukuthi i-terminal ingasebenzisa ngokuqondile i-ffmpeg. Ku-macOS ungayifaka ngomphathi wamaphakheji ovamile. Ngemva kokufaka, sebenzisa lo myalo wokuhlola:

ffmpeg -version

3. Lungiselela i-Whisper

Kule nqubo i-Whisper ayisetshenziselwa ukwenza umsindo, kodwa isebenza “njengesihloli somsindo”. Ihlola izingcezu zokuqabula noma ezingelona ulimi: uma ikwazi ukubhala ngokuzinzile amagama, ngokuvamile lokho kusho ukuthi imodeli empeleni ifunda amagama alingisa umsindo, esikhundleni sokukhiqiza umsindo ongakhulumi.

4. Yakha amafolda ezinto

Kuhle ukuhlukanisa amafayela kusenesikhathi ngokwemisebenzi yawo, okungenani kube nokuphumayo kwezwi, abazongenela umsindo wokuqabula, izinto zokuqabula eziphumelele ukuhlolwa, imisindo yendawo yoqobo, amafasitela omsindo wendawo asikiwe, kanye namafayela okugcina axutshiwe. Lokhu kugwema ukusebenzisa izinto ezingakahlolwa ngokuqondile emkhiqizweni wokugcina.

Isinyathelo sokuqala: Dala umbala wezwi wezilimi ezimbili

Sebenzisa incazelo ebhaliwe, ungakhethi kuphela emtatsheni wemibala yezwi

Ngokokuhlolwa kwe-README, imibala yezwi yesiShayina eseyikhona ingaba nesigqi sangaphandle esiNgisini, kanti ukuchaza ngokucacile isizinda sezilimi ezimbili nge-Voice Design kwenza kube lula ukuthola umphumela wemvelo ngesiShayina nangesiNgisi. Incazelo kufanele icacise izinhlobo ezine zolwazi: ikhono lolimi lwendabuko, ibanga lobudala, ubudlelwano noma ubuntu bomlingiswa, kanye nomuzwa ohlosiwe.

Nansi incazelo esetshenziswe enqubweni yokuqala; ingalungiswa ngokwezidingo zomlingiswa:

Native bilingual speaker of Mandarin Chinese (standard putonghua) and American English,
both at native level, with no cross-language accent in either. Male, 35–40. Studio quality.
Persona: mature, dominant, doting. Emotion: composed, doting, low-burning.
Rich, chesty, slightly gravelly low voice; unhurried, even pacing with weight on key words
and a breathy softness at the ends of sentences.

Umbhalo wokulalela akufanele ube yikhophi yokukhangisa. Khetha amazwi umlingiswa ebengawasho ngempela empilweni yakhe yansuku zonke, bese wenza abazongenela abaningi ngencazelo efanayo. Lapho ulalela, qala ngokuhlola ukuthi uzwakala yini njengomuntu ohlosiwe okhuluma ngokwemvelo, kunokuba uqhathanise kuphela ukuthi izwi lijule noma limnandi yini.

Khetha imodeli efanele

• Ingxoxo evamile: I-README incoma i-v3, efanele ukuveza imizwa yansuku zonke kokuqukethwe kombhalo.
• Izingcingo zesikhathi sangempela noma izigcawu ezisondelene: Kunconywa i-v4. Ifaneleka kakhulu ukuphatha isiShayina, isiNgisi nokuxubana kwalezi zilimi ngaphakathi kombala wezwi ofanayo, futhi ikwazi kangcono ukubona amathegi alula okuqondisa umsindo.
• Umbhalo omude wesiNgisi kuphela: Uma ungasebenzisi i-v4, ungakhetha i-eleven_turbo_v2_5, bese usetha ngokucacile language_code=en, ukuze unciphise indlela yesiShayina esiNgisini.

Lungisa amapharamitha wombala wezwi

Indawo yokuqala eyahlolwa ku-README imi kanje:

• stability=0.35: Ushintsho nedrama kubonakala kakhulu.
• stability=0.5: Kusondela kakhulu ekukhulumeni kwansuku zonke, futhi kunganciphisa umuzwa “wokulingisa ngamabomu”.
• similarity_boost=0.8: Isetshenziselwa ukugcina ukuvumelana kombala wezwi ohlosiwe.
• style=0.3: Amarekhodi okuqala athi i-v3 ayiyiboni le pharamitha, ngakho unganciki kuyo ukulawula i-v3.

Lapho ulungisa, shintsha into eyodwa kuphela ngesikhathi. Gcina umsindo oyisisekelo osuthole impendulo enhle, bese ulungisa ukuzinza noma ithegi eyodwa kuphela, ukuze ukwazi ukubona ngokunembile ukuthi ushintsho luvelaphi.

Isinyathelo sesibili: Hlanganisa ukuphefumula nesigqi sokukhuluma

Faka amathegi kuphela lapho kushintsha indlela yezwi

Ungafaki ithegi kuyo yonke imisho. Eqoqweni lemisho eyisithupha kuya kweyishumi, amathegi amabili noma amathathu alula ngokuvamile anele; imisho elandelayo ingaqhubeka nendlela yezwi yangaphambilini. Amagama avame ukusetshenziswa enqubweni yokuqala afaka low, close, quiet, soft kanye unhurried.

[low and close to her ear, unhurried]
……woza. Ungacashi.

……sondela kancane. Yebo.

[quiet laugh under the breath]
……usheshiselani.

Amagama anamandla afana no-intense, heavy, growl, strained kanye rough awavunyelwe ngokuphelele, kodwa ekuhlolweni kokuqala ayevame ukwenza izwi libe nehaba. Isu elithembeke kakhulu ukwehlisa, ukusondelisa nokwehlisa ijubane, kunokucela imodeli ukuthi idlale ngamandla.

Sebenzisa izimpawu zokuloba ukwakha ukuphefumula

Kunokuphindaphinda ukubhala heavy breathing, ukufaka “……” okusesilinganisweni ekuqaleni, ekugcineni noma phakathi komusho kuvame ukukhiqiza ukuma kwemvelo. Eqoqweni ngalinye hlela ukuphefumula okubili noma kathathu okugqamile, futhi uguqule indlela okubonakala ngayo, njengokukhipha umoya kancane, ukuma phakathi komusho, ukubamba umoya isikhashana, ukukhipha umoya ngokuhleka, noma ukunqamula umsindo phakathi.

Ukuphindaphinda uhlobo olufanayo lokuphefumula kulandelana kuveza umuzwa wokuphindaphinda. Ngemva kokukhiqiza, lalela yonke ingxenye, ungahloli umusho owodwa kuphela.

Yenza ingxenye ephelele ngesikhathi esisodwa

Lapho isigaba esiphelele senziwa ngesikhathi esisodwa, ukuphefumula, imizwa nobude bezwi kuba nokuqhubeka okungcono. Ukuhlanganisa imisho eyisishiyagalombili eyenziwe ngokwehlukana kuzokwenza ingxenye ngayinye izwakale sengathi iqala kabusha. Uma ingxoxo yesikhathi sangempela ingakwazi ukukhiqiza isigaba sonke, yenza ibhulokhi yemisho emibili noma cishe izinhlamvu zesiShayina ezingu-40, ugweme ibhulokhi yomusho owodwa.

Dlulisa umongo

I-v4 isekela previous_text kanye next_text. Lapho ukhiqiza ngamabhulokhi, dlulisela umbhalo wangaphambili nowangemuva kumodeli ukuze unciphise umuzwa wokufunda ngomshini. I-README igcizelela ukuthi i-v3 ayisekeli lezi zinkambu ezimbili; ukuzidlulisela ku-v3 kungadala iphutha, ngakho izicelo kufanele zakhiwe ngokwehlukana kuye ngemodeli.

Gwema umzila wolimi ongafanele

Uma uhlelo lunquma imodeli noma ulimi ngokwesilinganiso sezinhlamvu zesiShayina, kufanele luqale lususe amathegi esiNgisi angaphakathi kwabakaki abayisikwele. Kungenjalo, umusho omfushane wesiShayina ungahlulelwa ngephutha njengowesiNgisi ngenxa yamathegi amaningi esiNgisi. Uma usebenzisa i-v4, ungagcina imodeli efanayo ngokuqondile, unciphise ukushintsha phakathi kwesiShayina nesiNgisi.

Isinyathelo sesithathu: Lawula ijubane lokukhuluma

Ekuhlolweni kokuqala, ipharamitha ye-v4 ethi speed ayizange isebenze ngendlela ebilindelwe. Ukuze wenze i-v4 ihambe kancane, qala ngokwengeza izikhawu zemvelo embhalweni, bese usebenzisa i-ffmpeg ukwehlisa ijubane ngaphandle kokushintsha ubude bezwi:

ffmpeg -i input.mp3 -filter:a "atempo=0.9" output.mp3

Uma usebenzisa imodeli ye-turbo ngesiNgisi kuphela, indawo yokuqala esebenzayo erekhodwe ku-README ngu-speed=0.84, futhi ungangeza izikhawu phakathi kwemisho:

<break time="0.4s"/>

Ungaqoqi amathegi amaningi ezikhawu engxenyeni eyodwa. Ukuhlolwa kokuqala kuncoma cishe ayisithupha njengenani eliphezulu; amaningi kakhulu angenza umphumela ungazinzi.

Isinyathelo sesine: Yakha iqoqo lezinto zokuqabula ezingasetshenziswa futhi

Kungani kungafanele kubhalwe amagama alingisa umsindo

Ukuvumela imodeli yokuguqula umbhalo ibe yinkulumo ifunde amagama alingisa ukuqabula kuvame ukukhiqiza izinhlamvu noma amalunga abonakalayo, kunomsindo wemvelo ongakhulumi. Ukubhala kuphela amathegi ezenzo nakho kungaphenduka kube ukububula okuvamile. Ngakho iphrojekthi yokuqala yamukela isu leqoqo lezinto: “khiqiza ngobuningi kuqala, bese usika futhi uhlunge”.

Izinyathelo zokwenza

1. Sebenzisa umbala wezwi ofanayo nombhalo ofanayo oqukethe isenzo sokuqabula, ukhiqize ngokuphelele izikhathi ezinhlanu kuya kweziyisithupha. Inqubo yokuqala isebenzise i-v4 kanye stability=0.35.
2. Sebenzisa ukuthola ukuthula noma ithuluzi lokuhlela umsindo ukusika izingcezu ezingelona ulimi ezizokhethwa ekuphumeni ngakunye.
3. Dlulisa ucezu ngalunye ku-Whisper. Izingcezu ezikwazi ukubhalwa zibe izinhlamvu noma amagama acacile kufanele zilahlwe ngokuqondile.
4. Hlola isilinganiso samandla ahlukile namandla oqobo. Ekuhlolweni kokuqala, izingcezu ezithembeka kakhulu ngokuvamile zazingaphezu kuka-0.2; izingcezu ezazifundwe zaba amagama zazicishe zibe ngu-0.12 kuya ku-0.16; ukububula okuvamile kwakuvame ukuba ngaphansi kuka-0.1. Lezi zinombolo zivela ezintweni ezithile kuphela futhi zifaneleka njengereferensi yokuqala.
5. Lalela ngesandla izinto eziphumelele, bese uzihlela ngezigaba ezifana nokuqabula kancane, ukuqabula okujulile kanye nokushintsha kwendawo.
6. Lapho ubiza, khetha ngokungahleliwe ucezu olungasetshenziswanga muva nje, bese wenza ushintsho oluncane kubude bephimbo nejubane olungeqile cishe ku-3%. Inqubo yokuqala ivumela kuphela ukwehlisa ijubane, hhayi ukulisheshisa.

Ukuhlunga okuzenzakalelayo akukwazi ukuthatha indawo yokulalela ngesandla. Ku-README kwake kwavela ucezu olunomsindo oqhubekayo wokuhayiza; i-Whisper yayibhale into efana no-Sssss, kodwa ngenxa yokunganakwa kokuhlolwa lwangena emkhiqizweni wokugcina. Inqubo efanele kufanele ihlole umbhalo obhaliwe, izinkomba ze-waveform kanye nokuzwakala kwangempela ngesikhathi esisodwa.

Isinyathelo sesihlanu: Qedela ukuxuba kwamathrekhi amabili ngomsindo wangempela wendawo

Khetha izinto ezifanele

Inqubo yokuqala yaqhathanisa izindlela ezintathu: ukwenza imodeli yezwi ilingise, ukusebenzisa ukukhiqizwa kwemisindo nge-AI, kanye nokusebenzisa amarekhodi angempela. Ekugcineni kwakhethwa izinto zangempela ze-CC0 ezingenamazwi abantu nezinemvume ecacile. Ngaphambi kokusebenzisa noma yiliphi irekhodi langaphandle, kufanele kuqinisekiswe ilayisensi futhi kugcinwe umthombo.

Ungaphindaphindi ivolumu engaguquki ngokuqondile

Ukuzwakala okumaphakathi nokuphakama komsindo wokuqopha indawo kungahluka kakhulu, futhi kungaba nezindawo ezinde zokuthula. Ukusebenzisa isilinganiso sevolumu esingaguquki ngokuqondile kungadala ukuthi umsindo ungazwakali nhlobo noma uvale izwi. Indlela ezinzile yile:

1. Sika ngokungahleliwe amafasitela amabili entweni yoqobo, anobude obusondelene nobezwi.
2. Linganisa ukuphakama komsindo kwamafasitela womabili, bese ukhetha elinomsindo owusizo omningi.
3. Bala inzuzo ukuze ukuphakama komsindo wendawo kucishe kulingane nokuphakama kwamanje kwezwi kususwe ngo-12 dB.
4. Faka ukungena kancane kwemizuzwana engu-0.5 kumsindo wendawo.
5. Ngemva kokuphela kwezwi, gcina umsila womsindo cishe umzuzwana owodwa, bese usebenzisa ukuphuma kancane cishe kwemizuzwana engu-1.4.
6. Uma izinto zokuqabula sezivele zinokuthungwa okufanele, ungaphinde wengeze umsindo wendawo.

Isibonelo sokuxuba nge-ffmpeg

Cabanga ukuthi izwi lingu-v.mp3, iwindi lomsindo wendawo elisikiwe lingu-w.wav, GAIN iyinzuzo ebalwe ngemva kokuqondanisa ukuphakama komsindo, kanti END iyindawo lapho ukuphuma kancane kuqala khona:

ffmpeg -i v.mp3 -i w.wav -filter_complex \
 "[1:a]volume=GAIN,afade=t=in:st=0:d=0.5,afade=t=out:st=END-1.4:d=1.4[w];\
  [0:a]apad=pad_dur=1.0[v];[v][w]amix=inputs=2:duration=first:normalize=0[a]" \
 -map "[a]" -c:a libmp3lame -q:a 2 out.mp3

normalize=0 ibaluleke kakhulu. Ngokokuhlolwa kwe-README, uma uyishiya, i-amix ingase yehlise izwi kakhulu.

Hlukanisa amathegi okuxuba kumathegi okufundwa

Esigabeni sokubhala ungasebenzisa amathegi afana no-[low, close, wet], bese usebenzisa wet njengethegi yokuxuba yangemva kokukhiqiza. Ngaphambi kokucela imodeli yezwi, uhlelo kufanele lususe leli gama ukuze imodeli ingazami ukudlala umsindo; ngemva kokubuya kwezwi, izinto zangempela ezihambisanayo zingangezwa ngokwethegi. Uma ithegi elandelayo ingasayifaki leyo thegi, umsindo wendawo kufanele uphume kancane.

Isinyathelo sesithupha: Yenza ithrekhi yezwi ephelele

Hlukanisa umbhalo ube izinhlobo ezintathu zamabhulokhi: amabhulokhi amazwi ajwayelekile athunyelwe ekuguqulweni kombhalo abe yinkulumo; amabhulokhi angelona ulimi kuphela akhethwe eqoqweni lezinto; namabhulokhi anethegi yomsindo wendawo aqale akhiqizwe abe yizwi, bese exutshwa ngamathrekhi amabili.

Imisindo evela emithonjeni ehlukene ingaba namazinga esampula ahlukene, ngakho kufanele ahlanganiswe ngaphambi kokuhlanganisa, isibonelo ngokuyiguqulela ku-44100 Hz:

ffmpeg -i input.wav -ar 44100 output.wav

Ngemva kokuhlanganisa amazinga esampula, sebenzisa isihlungi se-concat se-ffmpeg ukuwahlanganisa abe yifayela elilodwa ngokulandelana kombhalo. Ngemva kokuqeda ukuhlanganisa, lalela kusukela ekuqaleni kuze kube sekupheleni, ugxile ekutheni umbala wezwi, umsindo ongemuva, umsindo ophelele nokuphefumula kushintsha ngokuzumayo yini phakathi kwezingxenye.

Isinyathelo sesikhombisa: Ukusetshenziswa ezingcingweni zesikhathi sangempela

Okubalulekile kumodi yesikhathi sangempela ukunciphisa isikhathi sokulinda kuyilapho kugcinwa umongo. Ungahlukanisa umbhalo ube izingxenye usebenzisa omaka abasemigqeni yabo, ukuze imisho elandelayo engenawo omaka abasha izuze izwi elikhona. Thumela ekukhiqizweni ngemva kokuqongelela imisho emibili noma cishe izinhlamvu zesiShayina ezingamashumi amane.

1. Ngesikhathi kudlalwa ingxenye yamanje, khiqiza ingxenye elandelayo kusengaphambili ngemuva.
2. Dlulisela ku-v4 umusho owandulele nowilandelayo wengxenye yamanje.
3. Izingxenye ezingenawo amazwi ngokuphelele zithathwe ngqo ebhokisini lezinsiza, ngaphandle kokucela ukuguqulwa kombhalo kube yinkulumo.
4. Vula imakrofoni kuphela ngemva kokuba kudlalwe umusho wokugcina, ukuze kugwenywe ukugqagqana kwemisindo.
5. Uma ukuhlanganiswa kwehluleka, bonisa umbhalo kuphela noma uthule isikhashana; ungashintsheli ngokuphuthuma kolunye uhlobo lwezwi.

Ukuphuka komlingiswa okubangelwa ukushintsha izwi phakathi nendawo kuvame ukubonakala kakhulu kunokuma cishe umzuzwana owodwa. Ngakho-ke, kufanele kugcinwe ukufana kwezwi lomlingiswa kuqala.

Amasu athuthukile okulungisa amaphutha

Yakha amasampula aqhathaniswayo angashintshi

Gcina umsindo oyisibonelo osuvele uqinisekisiwe ukuthi uzwakala ngokwemvelo. Ngaso sonke isikhathi shintsha into eyodwa kuphela phakathi kokuzinza, omaka, izimpawu zokuloba, imodeli noma isivinini sokucubungula ngemva kokukhiqiza, bese uwuqhathanisa nenguqulo eyisisekelo ngokulalela ungazi ukuthi iyiphi.

Hlukanisa omaka bokukhiqiza komaka bokucubungula ngemva kwalokho

Imodeli yezwi kufanele yamukele kuphela omaka bezwi okufanele iyilingise; umsindo wendawo, isigaba sezinsiza namandla okuxuba kungamamethadatha okucubungula ngemva kwalokho, futhi kufanele kususwe ngaphambi kwesicelo. Lokhu kugwema ukuthi imodeli ifunde imiyalelo yokukhiqiza noma izame ukulingisa imiphumela yomsindo.

Gcina umlando wokusetshenziswa kwakamuva

Gcina ulayini wokusetshenziswa kwakamuva webhokisi lezinsiza, isibonelo ugweme izingxenye eziyishumi ezisetshenziswe muva nje. Lokhu kunciphisa umuzwa wokuphindaphinda futhi kuzinza kakhulu kunokukhetha ngokungahleliwe ngokuphelele njalo.

Umkhawulo wekhwalithi ozisebenzayo

Ungangeza lezi zihloliwe emgqeni wokukhipha:

• Ingabe i-Whisper ibhale amagama okungafanele abe khona?
• Ingabe umehluko wamazinga aphezulu phakathi kwezwi nomsindo wendawo useduze nenani elihlosiwe?
• Ingabe izinga lesampula lihlanganisiwe?
• Ingabe umsindo wokugcina wemvelo ugcinwe ekupheleni komphumela?
• Ingabe imodeli esetshenziswe esicelweni iyawasekela amapharamitha adlulisiwe?
• Ingabe omaka besiNgisi basusiwe ngaphambi kokutholwa kolimi?

Izindleko, izimvume nemingcele yokuqukethwe

I-README ibhala ukuthi kusetshenziswa uhlelo lwe-Creator, olubiza ama-dollar ase-US angu-22 ngenyanga futhi lunomkhawulo wamagama acishe abe ngu-120,000; ukwakha izwi okokuqala kudla cishe izinhlamvu eziyizinkulungwane ezintathu, kanti ukusetshenziswa kwansuku zonke kusuka emakhulwini ambalwa kuya kweziyizinkulungwane ezimbili. I-README iphinde iveze ukuthi i-v4 ngaleso sikhathi yayinokunikezwa kwesaphulelo sesikhashana, ngakho intengo, umkhawulo nendlela yokukhokhisa yangempela kufanele kusekelwe ekhasini elisemthethweni ngesikhathi sokusetshenziswa.

Konke okuqukethwe kufanele kubhekisele kubantu abadala abacacile kuphela. Ungasebenzisi izilungiselelo zabalingiswa ezingaqondwa njengezabancane noma izinkulumo ezingacacisi ubudala. Inkundla izobuyekeza okuqukethwe ngokusekelwe embhalweni, futhi okuqukethwe okwephula imithetho kungaholela ekuvalweni kwe-akhawunti.

Lapho usebenzisa okuqoshiwe kwabanye abantu, kufanele uqinisekise imvume. Ngisho noma izinsiza zingalandwa mahhala, lokho akusho ukuthi kuvunyelwe ukuzisabalalisa kabusha noma ukuzisebenzisa kwezentengiselwano. Inqubo yokuqala isebenzisa kuphela izinsiza ze-CC0 futhi igcina umthombo wefayela ngalinye.

Isiphetho

Inkulumo esondelene ezwakala ngokwemvelo ayitholakali ngokunqwabelanisa omaka abanamandla. Indlela ethembeke kakhulu ukuvumela imodeli yezwi igxile emigqeni, emizweni elula nasekulinganiseni kokuphefumula, ihlukanise imisindo engenawo amazwi ibe yizinsiza ezibhalwe phansi futhi zahlolwa ngokulalelwa umuntu, bese isebenzisa i-ffmpeg ukwenza ukufanisa amazinga aphezulu, ukufiphala kokungena nokuphuma, kanye nokuhlanganisa.

Qala ngenguqulo encane: qala ngokwenza izwi elizinzile elinezilimi ezimbili, bese uqinisekisa ukukhiqizwa kwengxenye yonke nokudluliswa komongo, ngemva kwalokho wengeze ibhokisi lezinsiza nomsindo wendawo. Gcina amasampula ayisisekelo esinyathelweni ngasinye futhi ulungise into eyodwa kuphela ngesikhathi; lokhu kuvame ukwenza kube lula ukuthola imiphumela yemvelo, engaguquki futhi engasetshenziswa kabusha kunokwandisa omaka njalo.