Ingxelo ye Bug / Isicelo se Feature

Umkhangeli zincwadi we-AI Voice Deepfake

Fumana ukuthetha okuveliswe yi-AI, okuklonyiweyo, kunye nokudityaniswa kwelizwi nakweyiphi na iclip yesandi. Fumana inqaku lokulindeleka kunye nokulinganisa ukuthembeka kwimizuzu — i-inverse ye-TTS, eyenziwe kwimodeli efanayo.

Ubhaliso simahla

Asikho nasiphi na isandi se-TTS kwisiNgesi sakho. Nceda uncedo lwethu ukongeza isandi sakho! Intengiso yelizwi lakho

Layisha phezulu ifayile ye PDF

200 Iimpawu

Rhweba ngaphandle amanqaku encwadi ye Mozilla Khangela

Ubukhulu be-50MB kwifayili nganye. Imizuzwana ekuqala eyi-60 iphawulwe, iziqhoboshi ezide ziqhutywa.

— okanye urekhode kwi-microphone yakho —

00:00

Free — 10 ubhaqo ngosuku, akukho ubhaliso lufunekayo

Indlela esebenza ngayo

Isiqhoboshi siyi-HuBERT-base classifier (Meta AI, MIT licensed) elungelelaniswe kakuhle kwi-FoR, ASVspoof, kunye ne-WaveFake - ezintathu zezona zikhulu zabucala zokuthetha komuntu vs. ukuthetha okudityanisiweyo.

Ibuyisela ukwenzeka kweklipi eyenziwe yimodeli ye-TTS, i-voice clone, okanye enye inkqubo ye-AI, kunye nomgangatho woxanduva. Linganisa inqaku njengephawu olunamandla, hayi uqinisekiso olusemthethweni.

Usebenzisa njani le

Abashicileli abavavanya ushicilelo oluphucukileyo okanye olungaziwayo
abathengayo bendawo yokuthengwa kweempahla bevavanya ukuba iiseti zesandi
Iiqela zoxanduva nokhuseleko ziphawula ukuxhaphaza okuhlonisiweyo
Abaphandi ababeka uphawu lwedatha
Abavelisi bebonisa ukuba isandi sabo siyinyani

Iifomati Ezixhaswayo

MP3, WAV, FLAC, OGG, M4A

Ubukhulu be-50MB kwifayili nganye. Imizuzwana ekuqala eyi-60 iphawulwe, iziqhoboshi ezide ziqhutywa.

Iinkqubo zokufumana i-Deepfake

Qala ngokukhululekileyo, uphucule xa ufuna okuninzi

Iinketho zelizwe

10 ukufunyanwa ngosuku
Iiclip ukuya kuthi ga kwimizuzu engama-60
Inqaku lokungaqiniseki +
Akukho akhawunti ifunekayo

Ethandwa Kakhulu

I-akhawunti Ekhululekileyo

15,000 iimpawu ezikhululekileyo kwi-signup
200 iimpawu nganye yobhaqo
Ukukhangela imbali
Zonke iinkcukacha zemveliso (inqaku lohlobo olugqityiweyo)

Ubhaliso

I-Pro

Ufolo lwe-GPU oluphambili
Ubhaqo olukhulu nge-API
Ii-webhook zilumkisa nge-FAKE verdicts
Ulawulo lweeNkonzo

Yenza phezulu

Imibuzo ebuzwa rhoqo

It analyses an audio clip and tells you the probability that the speech is AI-generated, cloned, or otherwise synthetic. You get a 0-100% score plus a confidence rating, so you can decide whether the recording is trustworthy enough for publication, payment, or evidence.

I-HuBERT-base classifier (Meta AI, MIT licensed) elungelelaniswe kakuhle kwiintlobo ezintathu zezikhulu zabucala zokuthetha-no-ukuthetha-okwenziwa: FoR (Fake-or-Real), ASVspoof, kunye ne-WaveFake. Ilungelelaniso elifanelekileyo lishicilelwe njenge MelodyMachine/Deepfake-audio-detection-V2 kwi-Hugging Face.

Kwi uvavanyo lwabucala lwe FoR icwangcisa imodeli ephantsi ichaza ngaphezulu kwe 95% yobunjani, kwaye kwi ASVspoof- uhlobo lohlaselo ihlala ibonisa i TTS kunye nemveliso yobuhlobo belizwi. Ubunjani kwimodeli entsha umhleli akazange abone ngaphantsi - phatha inqaku njenge phawu olunamandla, hayi uphawu lwemithetho.

Ungakhuphela ii clip ukuya kwi 50MB kunye nemizuzu engama 60 yokuthetha. Ushicilelo olude luqhutywa ngokuzenzekelayo kwimizuzu yokuqala, eyaneleyo ukuba umhleli athathe isigqibo esizinzileyo.

Hayi. Iklip iqhubekeka kwimemori kwiseva yethu ye-GPU, isiphumo sibuyiselwa kuwe, kwaye uvimba weefayili olayishwe phezulu ucinywa ngokuzenzekelayo. Asizukungena, sizalise, okanye sisebenzise isandi sakho soqeqesho.

Ukukhangela kubiza iimpawu ezingama-200 ngeclip nganye - usebenziso oluncinci oluhlawulwayo kwinkqubo. Ii-akhawunti ezikhululekileyo zingaqhuba ukukhangela okumbalwa ngosuku; abasebenzisi ababhalisiweyo abahlawulwayo bangaqhuba ngokukhawuleza okupheleleyo nokufikelela kofolo oluphambili.

Uxanduva lubalulekile ngakumbi, hayi ngaphantsi, kwihlabathi eligcwele ilizwi eliguquliweyo. Intengiso ivumela abathengayo ukuba baqinisekise ukuba isampulu elayishwe ngumenzi lilungu lomuntu phambi kokuba bahlawule. Abashicileli banokujonga ukuthembeka koshicilelo olugqithisileyo. Iiqela zokhuseleko lweplagi zinokubona ukuxhaphaza okuklonyiweyo kwelizwi. I-detector iguqulwe kwi-TTS: i-expertise efanayo, injongo ephikisanayo.

Akukho nto - i-v1 detector ivelisa kuphela i-real vs synthetic. Unikezelo lwemodeli (injini ye-TTS eyenza le nto?) ikwindlela yokuhamba xa sinayo iimvavanyo ezilabelwe ngokwaneleyo ngemodeli nganye.

5.0/5 (1)

Ilungile ukuqalisa?

Ubhaliso simahla kwaye ufumane 15,000 iimpawu. Akukho khadi letyala lifunekayo.

Ubhaliso simahla Bonisa ixabiso

Umkhangeli zincwadi we-AI Voice Deepfake

Layisha phezulu ifayile ye PDF

Indlela esebenza ngayo

Usebenzisa njani le

Iifomati Ezixhaswayo

Iinkqubo zokufumana i-Deepfake

Imibuzo ebuzwa rhoqo

Yintoni eyenza i-deepfake detector?

Imodeli esebenza njani isixhobo sokukhangela?

Impumelelo ilungile kangakanani?

Ingaba isandi sam side kangakanani?

Ingaba isandi sam sigcinwa emva kokufunyanwa?

Ixabiso lithini?

Kutheni le nto ikhona kwiwebhusayithi ye-TTS?

Ingaba ikwazi ukuchaza ukuba yeyiphi imodeli ye-TTS evelise iclip yesandi?

Ilungile ukuqalisa?