Idanwo ti o sopọ mọ itọsọna · Lile Ipele

VALL-E ati Kodẹki Ede Models adanwo

Review VALL-E’s codec-token language-model approach, its staged token generation, and the conditions behind its reported few-shot voice results.

Jẹmọ itọsọna awọn ọnaVall E Ati Awọn awoṣe Ede Kodẹki
Ibeere 1 ti 8

Ero pataki wo ni o ṣe iyatọ VALL-E lati awọn eto ọrọ-si-ọrọ iṣaaju?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Ero pataki wo ni o ṣe iyatọ VALL-E lati awọn eto ọrọ-si-ọrọ iṣaaju?
    • O ṣe asọtẹlẹ awọn ami kodẹki nkankikan bi iṣẹ-ṣiṣe awoṣe ede dipo asọtẹlẹ spectrogram kan
    • O nlo sisẹ ifihan agbara ti ọwọ nikan
    • O nilo awọn wakati ohun afetigbọ lati ọdọ agbọrọsọ ibi-afẹde
    • O ṣe agbejade ọrọ lati ọrọ sisọ
  2. Elo ohun afetigbọ itọkasi VALL-E nilo lati ṣe oniye ohun agbọrọsọ tuntun kan?
    • Nipa wakati kan
    • Nipa iṣẹju-aaya mẹta
    • Ko si ohun itọkasi rara
    • Nipa ọgbọn iṣẹju
  3. Kodẹki nkankikan wo ni VALL-E lo lati ṣe agbejade awọn ami ohun afetigbọ rẹ?
    • MP3
    • Opus
    • EnCodec
    • WAV
  4. Kini idi ti VALL-E lo mejeeji autoregressive ati ipele ti kii ṣe adaṣe?
    • Lati tumọ ọrọ si awọn ede pupọ
    • Lati ya orin kuro ninu ọrọ sisọ
    • Lati funmorawon titẹ ọrọ
    • Nitori awọn ami kodẹki ti wa ni siwa, nitorinaa o ṣe asọtẹlẹ awọn ami isokuso lẹsẹsẹ ati awọn ami-ami ti o dara ni afiwe
  5. Ni aijọju bawo ni data ọrọ ọrọ VALL-E ṣe ikẹkọ lori, ti o mu ki ẹda oniye-shot odo lagbara?
    • Nipa awọn wakati 60,000
    • Nipa awọn wakati 100
    • Nipa awọn wakati 1,000
    • Nipa awọn wakati 10
  6. Yato si timbre, kini ohun miiran le VALL-E ṣe itọju lati agekuru itọkasi kukuru?
    • Awọn ọrọ gangan ti agbọrọsọ lailai
    • Ayika akositiki ti gbigbasilẹ
    • Fidio ti o tẹle e
    • Ọna kika faili atilẹba
  7. What does VALL-E condition on to imitate a target speaker in its demonstration setup?
    • A photograph and a written biography only
    • A long labeled transcript from every target-speaker recording
    • A short audio prompt from that speaker plus the text to speak
    • A speaker ID with no audio example
  8. What text representation does the VALL-E paper use as input to its text-to-code model?
    • A raw waveform of the target utterance
    • A speaker’s face embedding
    • A list of MIDI notes
    • A phoneme sequence