Ikibazo gihuza ibibazo · Hagati Urwego

Ikibazo cyihuta kandi kitari Autoregressive TTS Ikibazo

Kugenzura ibisa na ecran ya progaramu ya progaramu, kwaguka igihe, kwihuta kwihuta 2 hamwe ningufu zingufu, hamwe nubucuruzi bwimbitse.

Ikibazo 1 Bya 8

Bisobanura iki 'non-autoregressive' bisobanura murwego rwa FastSpeech?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Bisobanura iki 'non-autoregressive' bisobanura murwego rwa FastSpeech?
    • Itanga ama frame yose ya ecran muburyo bubangikanye aho gukurikiranya
    • Itanga amajwi yamajwi imwe murimwe ikurikiranye
    • Ntishobora gukora interuro ndende
    • Ntabwo bisaba amakuru yamahugurwa
  2. Ni ikihe kibazo cya moderi ya autoregressive nka Tacotron 2 yihuta cyane?
    • Kudashobora gukora amajwi ayo ari yo yose
    • Buhoro buhoro kandi wasimbutse cyangwa amagambo asubirwamo uhereye kumakosa yo kwitondera
    • Kubura inkunga yinyandiko yicyongereza
    • Gusaba ibyuma bya kwant
  3. Ni uruhe ruhare rw 'uburebure bugenzura' muri FastSpeech?
    • Ihagarika dosiye yanyuma y amajwi
    • Ikuraho guceceka kumajwi
    • Isubiramo buri fonema ihagarariwe ukurikije igihe cyateganijwe cyo guhuza uburebure bwamajwi
    • Ihindura fonema mu nyandiko
  4. Which additional speech variations does FastSpeech 2 condition on during training and inference?
    • Only speaker age and recording location
    • Video frames and subtitle color
    • A single fixed global loudness value
    • Pitch, energy, and duration information
  5. Ni ukubera iki igihe cyo kwihuta cya FastSpeech gikura gusa nuburebure bwinteruro?
    • Amakadiri yose yakozwe muburyo bubangikanye aho gukurikiranya
    • Irengagiza ibyinshi mubyanditswe
    • Gusa burigihe isohora clips imwe-isegonda
    • Ikoresha imbonerahamwe yo gushakisha interuro zanditswe mbere
  6. How does the length regulator use predicted durations in FastSpeech?
    • It selects the final audio codec
    • It expands phoneme-level representations to align with the target spectrogram length
    • It estimates the speaker’s identity from video
    • It converts text into a sequence of labels for translation
  7. What is one tradeoff of non-autoregressive spectrogram generation?
    • It guarantees identical prosody to an autoregressive model
    • It removes the need for a waveform vocoder
    • Parallel output can be faster, but duration and speech variation must be modeled explicitly
    • It cannot accept text as input
  8. What training target does FastSpeech 2 use instead of FastSpeech’s teacher-generated simplified outputs?
    • Only predicted durations with no acoustic target
    • A translated text transcript
    • The output of a GAN discriminator only
    • Ground-truth target speech representations