Idanwo ti o sopọ mọ itọsọna · Lile Ipele

HuBERT Ìdánwò Ọrọ Abojuto Ara-ẹni

Check how HuBERT creates cluster-derived targets for masked audio, iterates labels, and evaluates learned representations on speech benchmarks.

Jẹmọ itọsọna awọn ọnaỌrọ Abojuto Ara Hubert
Ibeere 1 ti 8

Bawo ni HuBERT ṣe ipilẹṣẹ awọn ibi-afẹde ti o gbiyanju lati ṣe asọtẹlẹ lakoko ikẹkọ?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Bawo ni HuBERT ṣe ipilẹṣẹ awọn ibi-afẹde ti o gbiyanju lati ṣe asọtẹlẹ lakoko ikẹkọ?
    • K-aisinipo tumọ si ikojọpọ ti awọn ẹya ohun sinu awọn ẹya ti o farapamọ
    • Awọn iwe afọwọkọ eniyan
    • Awọn aami ariwo laileto
    • Iyatọ lodi si awọn ayẹwo idamu
  2. Awoṣe ọrọ ti a mọ daradara wo ni o ṣe atilẹyin ero ikẹkọ asọtẹlẹ-boju ti HuBERT?
    • GPT-3
    • BERT
    • ResNet
    • Ọrọ2Vec
  3. Bawo ni HuBERT ṣe ilọsiwaju awọn aami ibi-afẹde rẹ lori awọn aṣetunṣe ikẹkọ atẹle?
    • O hires diẹ annotators
    • O ṣe afikun ariwo aise diẹ sii
    • O tun awọn iṣupọ nipa lilo awoṣe ti ara ti o ni ilọsiwaju awọn aṣoju inu inu
    • O di ikojọpọ akọkọ lailai
  4. Awọn ẹya wo ni a kojọpọ ni igbagbogbo ni aṣetunṣe akọkọ ti HuBERT?
    • Awọn ọrọ tiransikiripiti ipari
    • Awọn piksẹli aworan
    • Agbọrọsọ ID ifibọ
    • Awọn ẹya akositiki MFCC ti o rọrun
  5. Kini idi ti HuBERT le kọ ẹkọ eto ti o wulo paapaa nigbati iṣupọ rẹ jẹ alaipe?
    • O nilo nikan lati ṣe asọtẹlẹ awọn ID iṣupọ ni awọn ipo boju-boju, eyiti o duro iṣẹ ṣiṣe deede
    • O foju pa awọn iṣupọ patapata
    • O nlo akojọpọ ẹnu-ọna pipe
    • O ṣe akori eto idanwo naa
  6. Which pretraining challenge does HuBERT address by using offline cluster labels?
    • Speech has no temporal structure
    • There is no lexicon of sound units available during pretraining
    • Every speaker uses the same pitch
    • Audio contains no repeated patterns
  7. In the HuBERT paper, what relative WER reductions did the 1B-parameter model report on dev-other and test-other?
    • Exactly 50% on both sets
    • No measurable change on either set
    • Up to 19% on dev-other and 13% on test-other
    • Up to 13% on dev-other and 19% on test-other
  8. Which benchmark sets did the original HuBERT paper use for speech-recognition results?
    • ImageNet ati COCO
    • GLUE and SuperGLUE
    • SQuAD and TriviaQA
    • LibriSpeech and Libri-light