{"id":64732,"date":"2024-07-15T07:12:07","date_gmt":"2024-07-15T07:12:07","guid":{"rendered":"https:\/\/kamucalisani.net\/?p=64732"},"modified":"2024-07-15T07:12:07","modified_gmt":"2024-07-15T07:12:07","slug":"microsoft-vall-e-2-yapay-zeka-ses-taklidi-artik-ayirt-edilemez-seviyede","status":"publish","type":"post","link":"https:\/\/kamucalisani.net\/index.php\/2024\/07\/15\/microsoft-vall-e-2-yapay-zeka-ses-taklidi-artik-ayirt-edilemez-seviyede\/","title":{"rendered":"Microsoft VALL-E 2: Yapay zeka ses taklidi art\u0131k ay\u0131rt edilemez seviyede"},"content":{"rendered":"<p><figure> <span> <img decoding=\"async\" src=\"https:\/\/kamucalisani.net\/wp-content\/uploads\/2024\/07\/microsoft-vall-e-2-yapay-zeka-ses-taklidi-artik-ayirt-edilemez-seviyede-0-k9Qup1Tv.jpg\"\/> <\/span> Microsoft, ge\u00e7ti\u011fimiz y\u0131l\u0131n nisan ay\u0131nda insan seslerini taklit edebilen metinden konu\u015fmaya yapay zeka arac\u0131 VALL-E&#8217;yi tan\u0131tm\u0131\u015ft\u0131. O d\u00f6nemde VALL-E, \u00e7ok k\u0131sa bir ses \u00f6rne\u011finden sonra her t\u00fcrl\u00fc sesi taklit edebiliyordu. Ancak yeni duyurulan VALL-E 2, her sesi inan\u0131lmaz y\u00fcksek kalitede taklit edebiliyor. Bu y\u00fczden Microsoft, <strong>VALL-E 2<\/strong>&#8216;yi kamuoyuna sunulamayacak kadar ikna edici \u00f6rnekler \u00fcretti\u011fi i\u00e7in yay\u0131nlamama karar\u0131 ald\u0131. <\/figure>\n<p><b>Microsoft VALL-E 2 korkutuyor<\/b><\/p>\n<p>Daha \u00f6nce de <strong>metinden konu\u015fmaya <\/strong>(text-to-speech &#8211; TTS) yapay zeka ara\u00e7lar\u0131 g\u00f6rm\u00fc\u015ft\u00fck ancak VALL-E 2, ilk defa kar\u015f\u0131la\u015ft\u0131rma \u00f6l\u00e7\u00fctlerinde insanlarla ayn\u0131 seviyeye ula\u015fan t\u00fcr\u00fcn\u00fcn tek \u00f6rne\u011fi oluyor. Bu da modelin \u00e7ok ger\u00e7ek\u00e7i ses taklitleri yapabildi\u011fi anlam\u0131na geliyor. Microsoft&#8217;un VALL-E 2&#8217;yi halka a\u00e7\u0131k bir \u015fekilde yay\u0131nlamama nedeni de asl\u0131nda bu. A\u015fa\u011f\u0131daki ba\u011flant\u0131dan bir \u00f6rne\u011fe bakabilirsiniz. Ayr\u0131ca Microsoft&#8217;un kendi <strong><em>sitesindeki<\/em><\/strong> \u00f6rneklere de bakman\u0131z\u0131 tavsiye ederiz.<\/p>\n<p>VALL-E 2 ile tek bir ses dosyas\u0131yla yap\u0131lan ilk denemede modelin insan seviyesinde performans g\u00f6sterdi\u011fi belirtiliyor. Bununla birlikte VALL-E 2, karma\u015f\u0131kl\u0131\u011f\u0131 veya tekrar eden ifadeleri nedeniyle geleneksel olarak zor olan c\u00fcmlelerde bile konu\u015fma sentezini bozmuyor. VALL-E 2 esas\u0131nda ilk modelin \u00fczerine in\u015fa ediliyor ancak iki \u00f6nemi geli\u015ftirmeyle destekleniyor: &#8220;<strong>Tekrara Duyarl\u0131 \u00d6rnekleme<\/strong>&#8221; ve &#8220;<strong>Grupland\u0131r\u0131lm\u0131\u015f<\/strong> <strong>Kod<\/strong> <strong>Modelleme<\/strong>&#8220;.<\/p>\n<figure> <span> <img decoding=\"async\" src=\"https:\/\/kamucalisani.net\/wp-content\/uploads\/2024\/07\/microsoft-vall-e-2-yapay-zeka-ses-taklidi-artik-ayirt-edilemez-seviyede-1-3Ic78M8A.jpg\"\/> <\/span> \u0130lki, kod \u00e7\u00f6zme i\u015flemi s\u0131ras\u0131nda seslerin veya c\u00fcmlelerin sonsuz d\u00f6ng\u00fclerini \u00f6nleyen &#8220;belirte\u00e7lerin&#8221; (token) tekrarlar\u0131n\u0131 ele alarak yapay zekan\u0131n metni konu\u015fmaya d\u00f6n\u00fc\u015ft\u00fcrme \u015feklini geli\u015ftiriyor. Daha anla\u015f\u0131l\u0131r bir ifadeyle, bu \u00f6zellik VALL-E 2&#8217;nin konu\u015fma \u015feklini de\u011fi\u015ftirmeye yard\u0131mc\u0131 olarak <strong>daha ak\u0131c\u0131 ve do\u011fal<\/strong> g\u00f6r\u00fcnmesini sa\u011fl\u0131yor. <\/figure>\n<figure> <span> <img decoding=\"async\" src=\"https:\/\/kamucalisani.net\/wp-content\/uploads\/2024\/07\/microsoft-vall-e-2-yapay-zeka-ses-taklidi-artik-ayirt-edilemez-seviyede-2-nREGS1Jq.jpg\"\/> <\/span> Grupland\u0131r\u0131lm\u0131\u015f Kod Modelleme ise dizi uzunlu\u011funu ya da modelin tek bir giri\u015f dizisinde tek tek i\u015fledi\u011fi belirte\u00e7lerin say\u0131s\u0131n\u0131 azaltarak verimlili\u011fi art\u0131r\u0131yor. B\u00f6ylece VALL-E 2&#8217;nin <strong>konu\u015fma \u00fcretme h\u0131z\u0131 art\u0131r\u0131l\u0131yor<\/strong> ve uzun ses dosyalar\u0131 i\u015flenirken ortaya \u00e7\u0131kan zorluklar\u0131n \u00f6n\u00fcne ge\u00e7iliyor. <\/figure>\n<p>LibriSpeech ve VCTK veri k\u00fcmelerini kullanarak test edilen VALL-E 2 i\u00e7in ara\u015ft\u0131rmac\u0131lar, konu\u015fma sa\u011flaml\u0131\u011f\u0131, do\u011fall\u0131k ve konu\u015fma benzerli\u011fi a\u00e7\u0131s\u0131ndan \u00f6nceki TTS sistemlerinin geride b\u0131rak\u0131ld\u0131\u011f\u0131n\u0131 s\u00f6yledi.<\/p>\n<p>Microsoft, sahip oldu\u011fu yeteneklere ra\u011fmen potansiyel k\u00f6t\u00fcye kullan\u0131m riskleri nedeniyle VALL-E 2&#8217;yi halka sunmayacak. Ses klonlama ve deepfake teknolojisinin son derece eri\u015filebilir oldu\u011fu d\u00fc\u015f\u00fcn\u00fcld\u00fc\u011f\u00fcnde bu, yerinde bir karar. OpenAI gibi di\u011fer yapay zeka \u015firketleri de kendi ses teknolojilerine benzer k\u0131s\u0131tlamalar uyguluyor.<\/p>\n\n<p>Kaynak\u00a0 :\u00a0<span style=\"background-color: rgb(255, 249, 236); color: rgb(55, 58, 60); font-size: 14px;\">https:\/\/www.donanimhaber.com\/vall-e-2-ile-yapay-zeka-ses-taklidi-artik-ayirt-edilemez-seviyede&#8211;179338<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Microsoft, ge\u00e7ti\u011fimiz y\u0131l\u0131n nisan ay\u0131nda insan seslerini taklit edebilen metinden konu\u015fmaya yapay zeka arac\u0131 VALL-E&#8217;yi tan\u0131tm\u0131\u015ft\u0131. O d\u00f6nemde VALL-E, \u00e7ok k\u0131sa bir ses \u00f6rne\u011finden sonra her t\u00fcrl\u00fc sesi taklit edebiliyordu. Ancak yeni duyurulan VALL-E &#8230;<\/p>\n","protected":false},"author":1,"featured_media":64733,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[3277,555,228,506],"class_list":["post-64732","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","tag-konusma","tag-microsoft","tag-ses","tag-tek"],"_links":{"self":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/64732","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/comments?post=64732"}],"version-history":[{"count":1,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/64732\/revisions"}],"predecessor-version":[{"id":64737,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/posts\/64732\/revisions\/64737"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/media\/64733"}],"wp:attachment":[{"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/media?parent=64732"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/categories?post=64732"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kamucalisani.net\/index.php\/wp-json\/wp\/v2\/tags?post=64732"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}