Mozilla Common Voice 8.0 Voice Update

Mozilla yakaburitsa gadziriso kune yayo Common Voice dhataseti, iyo inosanganisira mataurirwo emasampuli kubva kuvanhu vangangosvika mazana maviri ezviuru. Iyo data inoburitswa seruzhinji domain (CC200). Iwo akarongwa seti anogona kushandiswa mumashini ekudzidza masisitimu kuvaka kucherechedzwa kwekutaura uye mamodheru. Kuenzaniswa neyakagadziridzwa yapfuura, huwandu hwemashoko ekutaura muunganidzwa hwakawedzera ne0% - kubva pa30 kusvika 13.9 zviuru maawa ekutaura. Huwandu hwemitauro inotsigirwa hwakawedzera kubva pa18.2 kusvika ku67.

Iyo yakatarwa yemutauro weRussia inovhara 2452 vatori vechikamu uye 193 maawa ekutaura zvinhu (kwaiva nevatori vechikamu 2136 nemaawa 173), kumutauro weBelarusian - vatori vechikamu 6160 uye maawa 987 (paiva nevatori vechikamu 3831 uye maawa 356), nokuda kwemutauro weUkraine - 684 vatori vechikamu uye maawa 76 (paiva nevatori vechikamu 615 uye maawa 66). Kupfuura 79 zviuru zvevanhu vakatora chikamu mukugadzirira kwezvinhu muChirungu, vachiraira maawa e2886 ekutaura kwakasimbiswa (paiva nevatori vechikamu zviuru makumi manomwe neshanu uye maawa 75).

Ngatikuyeuchidzei kuti Common Voice project inotarisirwa kuronga basa rekubatana kuti riunganidze dhatabhesi yemaitiro ezwi iyo inofunga nezvekusiyana kwemazwi uye maitiro ekutaura. Vashandisi vanokokwa kutaura mazwi anoratidzwa pachiratidziro kana kuongorora mhando yedata rakawedzerwa nevamwe vashandisi. Iyo yakaunganidzwa dhatabhesi ine marekodhi emataurirwo akasiyana eakajairwa mitsara yekutaura kwevanhu inogona kushandiswa pasina kurambidzwa mumashini ekudzidza masisitimu uye mumapurojekiti ekutsvagisa. Sekureva kwemunyori weVosk inoenderera mberi yekuzivisa matauriro raibhurari, izvo zvisingabatsiri zveCommon Voice seti ndeye divi rimwe chete remazwi ezwi (kunyanya kwevanhu vechirume 20-30 makore, uye kushomeka kwezvinhu nemanzwi evakadzi. , vana nevakura), kushaikwa kwekusiyana-siyana muduramazwi (kudzokororwa kwezvirevo zvakafanana) uye kugoverwa kwezvakarekodhwa mumhando yeMP3 inoshatisa.

Pamusoro pezvo, tinogona kucherechedza kuburitswa kweNVIDIA NeMo 1.6 toolkit, iyo inopa nzira dzekudzidza muchina dzekugadzira masisitimu ekuzivikanwa kwekutaura, kusanganisa kwekutaura uye kugadzirisa mutauro chaiwo. NeMo inosanganisira akagadzirira-kushandisa-akadzidziswa mamodhiyo emuchina ekudzidza masisitimu anoenderana nePyTorch chimiro, chakagadzirirwa neNVIDIA vachishandisa Common Voice matauriro data uye inovhara mitauro yakasiyana-siyana, mataurirwo uye matauriro. Iwo mamodheru anogona kubatsira kune vaongorori vanogadzira izwi-based dialogue system, transcription platforms, uye automated call centers. Semuenzaniso, NVIDIA NeMo inoshandiswa mune otomatiki izwi masevhisi eMTS uye Sberbank. Iyo NeMo kodhi yakanyorwa muPython uchishandisa PyTorch uye yakagoverwa pasi peiyo Apache 2.0 rezinesi.

Source: opennet.ru

Voeg