![\"\"](\"https:\/\/ekonomigundemi.com.tr\/wp-content\/uploads\/2026\/10\/91300f6d62ddfb84bdedc78277a8ad2e_1.jpg\")
\n<\/p>\nTether AI Research taraf\u0131ndan geli\u015ftirilen QVAC TranscribePsy, ak\u0131ll\u0131 telefonlar, ak\u0131ll\u0131 g\u00f6zl\u00fckler ve insans\u0131 robotlarda tamamen cihaz \u00fczerinde \u00e7al\u0131\u015fmak \u00fczere tasarland\u0131. Model, ses kay\u0131tlar\u0131n\u0131 bulut sunucular\u0131na g\u00f6ndermeden \u00e7evrimd\u0131\u015f\u0131 olarak i\u015fleyebiliyor. Bluetooth veya Wi-Fi \u00fczerinden ba\u011flanan ak\u0131ll\u0131 g\u00f6zl\u00fcklerden gelen sesleri ger\u00e7ek zamana yak\u0131n bi\u00e7imde i\u015fleyebilen sistem, insans\u0131 robotlar\u0131n kendi i\u015flemcilerinde de \u00e7al\u0131\u015fabiliyor. Sesle \u00e7al\u0131\u015fan teknolojilerin evlerde, i\u015f yerlerinde ve g\u00fcnl\u00fck ya\u015famda yayg\u0131nla\u015fmas\u0131yla birlikte model, \u00f6zel konu\u015fmalar\u0131n ve hassas bilgilerin buluta aktar\u0131lmas\u0131 ihtiyac\u0131n\u0131 ortadan kald\u0131rmay\u0131 hedefliyor. B\u00f6ylece kullan\u0131c\u0131lar, ses verileri \u00fczerindeki kontrollerini koruyabiliyor.<\/p>\n
Kompakt yap\u0131s\u0131yla y\u00fcksek performans<\/p>\n
Yaln\u0131zca 254 milyon parametreye sahip TranscribePsy, do\u011fruluk, h\u0131z ve bellek verimlili\u011fi alanlar\u0131nda \u00f6ne \u00e7\u0131k\u0131yor. Model, Open ASR Leaderboard\u2019da y\u00fczde 8,02, Multilingual LibriSpeech\u2019te y\u00fczde 6,58 ve Common Voice\u2019ta y\u00fczde 21,56 kelime hata oran\u0131 (WER) elde etti. Multilingual LibriSpeech testinde kendisinden \u00fc\u00e7 ila d\u00f6rt kat daha b\u00fcy\u00fck baz\u0131 modelleri geride b\u0131rakt\u0131.<\/p>\n
Alt\u0131 ak\u0131ll\u0131 telefonda yap\u0131lan testlerde en h\u0131zl\u0131 model olan TranscribePsy, i\u015flemci performans\u0131nda kendisinden sonra gelen modele k\u0131yasla 2,6 kat, Whisper-small modeline k\u0131yasla ise 5,7 kat daha h\u0131zl\u0131 \u00e7al\u0131\u015ft\u0131. Apple Core ML s\u00fcr\u00fcm\u00fc, iPhone 17 Pro Max\u2019te 10 dakikal\u0131k sesi 1,65 saniyede yaz\u0131ya d\u00f6n\u00fc\u015ft\u00fcr\u00fcrken test edilen rakip Core ML modeline k\u0131yasla disk alan\u0131n\u0131n yar\u0131s\u0131ndan az\u0131n\u0131 ve 1,7 kat daha az belle\u011fi kulland\u0131.<\/p>\n
Model, diz\u00fcst\u00fc bilgisayarlarda 25 Avrupa dilini, telefonlarda ise 8 Avrupa dilini destekliyor. Ayr\u0131ca iPhone\u2019un Neural Engine birimi i\u00e7in optimize edilmi\u015f \u00f6zel bir s\u00fcr\u00fcm sunuyor.<\/p>\n
Ses odakl\u0131 yapay zek\u00e2 uygulamalar\u0131na altyap\u0131<\/p>\n
TranscribePsy, s\u00fcrekli ve ger\u00e7ek zamana yak\u0131n ses i\u015fleme i\u00e7in geli\u015ftirildi. Cihaz \u00fczerinde \u00e7al\u0131\u015fan ses tan\u0131ma teknolojisi, bulut tabanl\u0131 i\u015flemeyle ili\u015fkili gecikmeleri ve veri s\u0131z\u0131nt\u0131s\u0131 risklerini azaltabiliyor. Bu yakla\u015f\u0131m; t\u0131bbi g\u00f6r\u00fc\u015fmeler, hukuki m\u00fclakatlar, terapi seanslar\u0131 ve gizlilik gerektiren gazetecilik faaliyetleri gibi hassas alanlarda \u00f6nem ta\u015f\u0131yor. \u0130nsans\u0131 robotlarda ise kompakt model, g\u00f6rsel alg\u0131lama, mek\u00e2nsal haritalama ve fiziksel kontrol gibi di\u011fer g\u00f6revler i\u00e7in i\u015flem kaynaklar\u0131n\u0131n korunmas\u0131na yard\u0131mc\u0131 oluyor.<\/p>\n
Yay\u0131mlanan paket; 25 Avrupa dilini destekleyen diz\u00fcst\u00fc bilgisayar modelini, sekiz temel dile odaklanan alternatif s\u00fcr\u00fcm\u00fc, qvac-fabric-speech\u2019in parakeet.cpp motorunda \u00e7al\u0131\u015fan GGUF mobil s\u00fcr\u00fcm\u00fcn\u00fc ve iPhone Neural Engine i\u00e7in optimize edilen Core ML s\u00fcr\u00fcm\u00fcn\u00fc i\u00e7eriyor. QVAC ekosisteminde TranscribePsy, VisionPsy, TranslatePsy-Nano ve MedPsy ile olas\u0131 entegrasyonlar i\u00e7in ses-metin d\u00f6n\u00fc\u015f\u00fcm altyap\u0131s\u0131 sunuyor. Bu yeteneklerin bir araya gelmesi; sesle belge analizini, \u00e7eviriyi, navigasyonu ve \u00f6zel not almay\u0131 destekleyebilecek uygulamalar\u0131n \u00f6n\u00fcn\u00fc a\u00e7\u0131yor.<\/p>\n
\n<\/p>\n
\u201cYapay zek\u00e2n\u0131n gelece\u011fi, ki\u015fisel ya\u015fam\u0131m\u0131z\u0131 merkezi platformlar\u0131n eri\u015fimine a\u00e7mam\u0131z\u0131 gerektirmemeli\u201d<\/p>\n
Tether CEO\u2019su Paolo Ardoino, TranscribePsy\u2019\u0131n geli\u015fmi\u015f sesli yapay zek\u00e2n\u0131n h\u0131zl\u0131, yetenekli ve tamamen cihaz \u00fczerinde \u00e7al\u0131\u015fabilece\u011fini g\u00f6sterdi\u011fini belirtti. Ardoino, yapay zek\u00e2n\u0131n telefonlara, g\u00f6zl\u00fcklere ve makinelere entegre olaca\u011f\u0131n\u0131 ancak bunun kullan\u0131c\u0131 verilerinin bu cihazlar\u0131n d\u0131\u015f\u0131na \u00e7\u0131kmas\u0131n\u0131 gerektirmedi\u011fini vurgulayarak, kullan\u0131c\u0131lar\u0131n verileri ve gizlilikleri \u00fczerindeki kontrol\u00fcn\u00fc koruyan \u00e7\u00f6z\u00fcmler geli\u015ftirmeyi hedeflediklerini ifade etti.<\/p>\n
A\u00e7\u0131k kaynakl\u0131 ve yeniden \u00fcretilebilir<\/p>\n
TranscribePsy, ara\u015ft\u0131rma ve e\u011fitim ama\u00e7lar\u0131 i\u00e7in Apache 2.0 lisans\u0131 alt\u0131nda yay\u0131mland\u0131. A\u00e7\u0131k model a\u011f\u0131rl\u0131klar\u0131, GGUF mobil s\u00fcr\u00fcmleri, Apple Silicon Core ML derlemeleri ve \u015feffaf, tekrarlanabilir performans kar\u015f\u0131la\u015ft\u0131rmalar\u0131na y\u00f6nelik eksiksiz de\u011ferlendirme yap\u0131land\u0131rmalar\u0131 da pakete d\u00e2hil edildi. Model, Tether AI Research\u2019\u00fcn ses, metin, g\u00f6r\u00fcnt\u00fc ve videoyu do\u011frudan u\u00e7 cihazlarda i\u015fleyebilen kompakt ve verimli yapay zek\u00e2 modelleri geli\u015ftirme \u00e7al\u0131\u015fmalar\u0131n\u0131n bir par\u00e7as\u0131. Uzun vadeli hedef, hassas verilerin cihaz d\u0131\u015f\u0131na \u00e7\u0131kmas\u0131n\u0131 gerektirmeden geli\u015fmi\u015f \u00e7ok modlu yapay zek\u00e2y\u0131 ihtiya\u00e7 duyulan her yerde kullan\u0131labilir h\u00e2le getirmek.<\/p>\n
QVAC hakk\u0131nda<\/p>\n
QVAC, a\u00e7\u0131k, merkeziyetsiz ve uyarlanabilir zek\u00e2 sistemleri geli\u015ftirmeye odaklanan Tether Data\u2019n\u0131n yapay zek\u00e2 ara\u015ft\u0131rma giri\u015fimidir. \u201cYerel Yapay Zek\u00e2, S\u0131n\u0131rs\u0131z Zek\u00e2\u201d misyonuyla hareket eden QVAC, yapay zek\u00e2n\u0131n kurumsal veri merkezlerinde yo\u011funla\u015fmak yerine her cihazda \u00e7al\u0131\u015fabildi\u011fi ve \u00f6\u011frenebildi\u011fi bir yakla\u015f\u0131m\u0131 destekliyor.<\/p>\n
Tether Data hakk\u0131nda<\/p>\n
Tether Data, S.A. de C.V., Tether\u2019\u0131n teknoloji arac\u0131l\u0131\u011f\u0131yla \u00f6zg\u00fcrl\u00fck, \u015feffafl\u0131k ve inovasyonu geli\u015ftirme vizyonunun bir par\u00e7as\u0131d\u0131r. \u015eirket, bireylerin ve kurumlar\u0131n gereksiz arac\u0131lar olmadan do\u011frudan ba\u011flant\u0131 kurmas\u0131n\u0131 ve bilgi payla\u015fmas\u0131n\u0131 sa\u011flamay\u0131 ama\u00e7l\u0131yor. G\u00fcvenli, e\u015fler aras\u0131 sistemler arac\u0131l\u0131\u011f\u0131yla kullan\u0131c\u0131lar\u0131n verileri, ileti\u015fimleri ve dijital etkile\u015fimleri \u00fczerindeki kontrol\u00fcn\u00fc art\u0131rmay\u0131 hedefleyen Tether Data, merkezi modellerin yerine gizlilik, verimlilik ve dayan\u0131kl\u0131l\u0131k odakl\u0131 merkeziyetsiz altyap\u0131lar geli\u015ftirmeye odaklan\u0131yor.<\/p>\n
<\/strong> Tether AI Research, bulut i\u015flemeye ihtiya\u00e7 duymadan do\u011frudan cihaz \u00fczerinde \u00e7al\u0131\u015fan 254 milyon parametreli \u00e7ok dilli ses tan\u0131ma modeli QVAC TranscribePsy\u2019\u0131 kullan\u0131ma sundu. iPhone 17 Pro Max\u2019te 10 dakikal\u0131k sesi 1,65 saniyede yaz\u0131ya d\u00f6n\u00fc\u015ft\u00fcrebilen model, kullan\u0131c\u0131 verilerinin gizlili\u011fini korurken sesle etkile\u015fimli yapay zek\u00e2 uygulamalar\u0131n\u0131n geli\u015ftirilmesini destekliyor.<\/p>\n","protected":false},"author":1,"featured_media":167946,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"googlesitekit_rrm_CAow07ekDA:productID":"","footnotes":""},"categories":[13,9],"tags":[],"class_list":["post-167945","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","category-manset"],"_links":{"self":[{"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/posts\/167945","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/comments?post=167945"}],"version-history":[{"count":0,"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/posts\/167945\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/media\/167946"}],"wp:attachment":[{"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/media?parent=167945"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/categories?post=167945"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ekonomigundemi.com.tr\/index.php\/wp-json\/wp\/v2\/tags?post=167945"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\nEKONOM\u0130 GAZETES\u0130 <\/p>\n","protected":false},"excerpt":{"rendered":"