{"id":119595,"date":"2024-08-23T15:43:08","date_gmt":"2024-08-23T13:43:09","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/mozilla-razvivaet-instrumentarij-dlya-raspoznavaniya-rechi-whisperfile"},"modified":"2024-08-23T15:43:08","modified_gmt":"2024-08-23T13:43:09","slug":"mozilla-razvivaet-instrumentarij-dlya-raspoznavaniya-rechi-whisperfile","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/news\/mozilla-razvivaet-instrumentarij-dlya-raspoznavaniya-rechi-whisperfile","title":{"rendered":"Mozilla po zhvillon mjetin Whisperfile p\u00ebr njohjen e t\u00eb folurit","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Kompania Mozilla po zhvillon mjetin e njohjes s\u00eb z\u00ebrit Whisperfile, duke p\u00ebrfshir\u00eb nj\u00eb implementim t\u00eb pavarur me performanc\u00eb t\u00eb lart\u00eb t\u00eb modelit t\u00eb m\u00ebsimit t\u00eb makinerive Whisper, i zhvilluar dhe i hapur nga organizata OpenAI. Mjeti \u00ebsht\u00eb krijuar mbi baz\u00ebn e whisper.cpp, nj\u00eb implementim i modelit Whisper n\u00eb C\/C++, i krijuar nga Georgi Gerganov (autori i llama.cpp). Kodi \u00ebsht\u00eb shkruar n\u00eb gjuh\u00ebn C++ dhe shp\u00ebrndahet n\u00ebn licenc\u00ebn MIT. <\/p>\n
Whisperfile po zhvillohet nga ekipi Mozilla Ocho dhe plot\u00ebson projektin llamafile, i destinuar p\u00ebr t\u00eb krijuar skedar\u00eb ekzekutues universale p\u00ebr t\u00eb drejtuar modele t\u00eb m\u00ebdha gjuh\u00ebsore t\u00eb m\u00ebsimit t\u00eb makinave (LLM). N\u00eb p\u00ebrputhje me llamafile, projekti whisperfile lejon q\u00eb n\u00eb baz\u00eb t\u00eb nj\u00eb skedari me parametra t\u00eb modelit t\u00eb m\u00ebsimit t\u00eb makinave n\u00eb formatin GGUF t\u00eb gjenerohet nj\u00eb skedar ekzekutiv, i cili mund t\u00eb ekzekutohet n\u00eb sisteme t\u00eb ndryshme operative mbi pajisje me procesor\u00eb AMD64 dhe ARM64. Kodi i kompiluar mund t\u00eb lidhet me bibliotek\u00ebn standarde C t\u00eb Cosmopolitan, e cila ofron mund\u00ebsin\u00eb p\u00ebr t\u00eb krijuar p\u00ebrmbledhje aplikacionesh q\u00eb mund t\u00eb ekzekutohen n\u00eb Linux, FreeBSD, macOS, OpenBSD, NetBSD dhe Windows. <\/p>\n
Kur ekzekutohet skedari ekzekutiv, si paramet\u00ebr hyr\u00ebs d\u00ebrgohet nj\u00eb skedar me z\u00eb t\u00eb folur n\u00eb formatin wav, mp3, ogg ose flac, dhe n\u00eb daljen e tij ruhet teksti i njohur. N\u00eb praktik\u00eb, projekti mund t\u00eb p\u00ebrdoret p\u00ebr t\u00eb zgjidhur detyra si gjenerimi i titujve tekstual\u00eb p\u00ebr video, krijimi i nj\u00eb logu p\u00ebr thirrjet vokale dhe video, shnd\u00ebrrimi i materialeve t\u00eb regjistruara t\u00eb z\u00ebrit n\u00eb tekst, organizimi i hyrjes me z\u00eb. Me ndihm\u00ebn e Whisperfile, k\u00ebto detyra mund t\u00eb zgjidhen n\u00eb nj\u00eb sistem lokal pa u drejtuar n\u00eb sh\u00ebrbime t\u00eb jashtme. <\/p>\n
P\u00ebr m\u00eb tep\u00ebr, mb\u00ebshtetet puna si nj\u00eb server HTTP q\u00eb p\u00ebrpunon k\u00ebrkesat p\u00ebr njohjen e z\u00ebrit p\u00ebrmes Web API-s\u00eb. P\u00ebr t\u00eb p\u00ebrshpejtuar pun\u00ebn me modelin, mund t\u00eb angazhohen GPU-t\u00eb dhe instrukcionet AVX. Mjeti gjithashtu mund t\u00eb jap\u00eb koeficient\u00eb besueshm\u00ebrie, duke lejuar ngjyrimin e fjal\u00ebve t\u00eb njohura n\u00eb var\u00ebsi t\u00eb sakt\u00ebsis\u00eb s\u00eb p\u00ebrcaktimit t\u00eb tyre.
<\/center> <\/p>\n
Modeli i p\u00ebrdorur Whisper \u00ebsht\u00eb trajnuar me 680,000 or\u00eb t\u00eb dh\u00ebnash t\u00eb fjal\u00ebve, duke p\u00ebrfshir\u00eb fusha t\u00eb ndryshme tematike dhe gjuh\u00eb (2\/3 e t\u00eb dh\u00ebnave jan\u00eb n\u00eb anglisht). Modeli p\u00ebrballon mir\u00eb njohjen e fjal\u00ebs me akcent, identifikon argot teknik, mb\u00ebshtet p\u00ebrcaktimin automatik t\u00eb gjuh\u00ebs dhe mund t\u00eb funksionoj\u00eb edhe n\u00eb prani t\u00eb zhurm\u00ebs n\u00eb sfond. P\u00ebr t\u00eb folur n\u00eb anglisht, sistemi tregon nj\u00eb nivel besueshm\u00ebrie dhe sakt\u00ebsie t\u00eb njohjes automatikisht q\u00eb \u00ebsht\u00eb af\u00ebr njohjes nga njeriu. P\u00ebrve\u00e7 transkriptimit t\u00eb z\u00ebrit n\u00eb tekst, modeli mund t\u00eb p\u00ebrdoret gjithashtu p\u00ebr p\u00ebrkthimin e fjal\u00ebs n\u00eb nj\u00eb gjuh\u00eb tjet\u00ebr. \n Burimi: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"
![\"Mozilla](\"\/wp-content\/uploads\/2024\/08\/f4c916bc3a9281c8e5016090f183fada.png\")
<\/center> <\/p>\n