Kaddu yu Gaaw ak Quiz TTS budul dellu ginaaw

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

1. Luy 'budul dellu ginaaw' ci wàllu FastSpeech?
   • Dafay defar kadre spectrogram yépp ci paralel te baña toppalante
   • Dafay defar ay kadre audio benn-benn ci toppalante
   • Mënul jëfandikoo frase yu gudd
   • Soxlawolen ay done tàggat
2. Ban jafe-jafe ci xeetu dellu ginaaw bu melni Tacotron 2 la FastSpeech gëna wax ci?
   • Mënul defar benn audio
   • Generation bu yeex ak tëb wala baamtu kàddu yu bawoo ci njuumti bàyyi xel
   • Ñàkum ndimbal ci mbindu angale
   • Dafay laaj jumtukaayi kwantum
3. Lan mooy wareefu 'régulatëru guddaay' ci FastSpeech?
   • Dafay kompresse fichier audio bi mujj
   • Dafay dindi noppi ci enregistrement yi
   • Dafay baamtu fonem bu nekk ci diir biñu ko séentu ngir méngoo ak guddaayu audio bi
   • Dafay tekki fonem yi ci mbind
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. Lan moo waral diiru inference bu FastSpeech di yokk bu baax ak guddaayu frase bi?
   • Kadre yépp ñu ngi leen di defar ci paralel du ñu toppalante
   • Dafay baña bàyyi xel ci mbind yiñ dugal
   • Benn clip rek lay génne
   • Dafay jëfandikoo tablo seetlu frase yuñ njëkka enregistre
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