Maswali yanayounganishwa na mwongozo · Kati Kiwango

Maswali ya MuzikiGen

Kagua maandishi na hali ya muziki ya MusicGen, tokeni za EnCodec, uingiliaji wa mtiririko wa moja kwa moja, na vikwazo vya utumiaji wa modeli.

Njia za mwongozo zinazohusianaMuziki
Swali 1 ya 8

Which input can a melody-conditioned MusicGen variant accept in addition to text?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. Which input can a melody-conditioned MusicGen variant accept in addition to text?
    • A reference melody
    • Picha na nambari ya tempo
    • Kitanzi cha ngoma na klipu ya video
    • Lebo ya aina pekee
  2. Ni kodeki gani ya neural hutoa tokeni za sauti ambazo MusicGen inatabiri?
    • SoundStream
    • EnCodeki
    • MP3
    • Whisper
  3. How does MusicGen model multiple EnCodec code streams?
    • Inaepuka kutumia mtandao wowote wa neva
    • Inahitaji muundo tofauti kwa sekunde ya sauti
    • A single-stage autoregressive Transformer uses delayed, interleaved streams
    • Inatoa MIDI pekee
  4. Je, MusicGen inafuataje wimbo uliotolewa bila kunakili rekodi halisi?
    • Inakariri faili ya sauti
    • Huiandika kwenye karatasi ya muziki kwanza
    • Inapuuza melody kabisa
    • Inatumia chromagram (wasifu wa darasa la lami) kwa urekebishaji
  5. Ni mradi na kampuni gani iliyotolewa MusicGen?
    • Ufundi wa Sauti wa Meta
    • SautiLM ya Google
    • Jukebox ya OpenAI
    • Utulivu wa Sauti ya AI
  6. Ni nini hutoa hali ya maandishi ya MusicGen?
    • Mtandao wa picha wa kushawishi
    • Kisimbaji maandishi cha T5
    • Jenereta ya nambari isiyo ya kawaida
    • Avkodare ya EnCodec
  7. What maximum duration does the standard AudioCraft MusicGen implementation set for a single generation?
    • Exactly 12 seconds with no extension support
    • Two minutes in every configuration
    • Up to 30 seconds before using its extended-generation path
    • Unlimited duration in a single model call
  8. What license does the official MusicGen model card list for its pretrained weights?
    • Apache 2.0 for all weights and code
    • Public domain with no conditions
    • MIT for all weights and training data
    • CC-BY-NC 4.0