Hadal Degdega iyo Kediska TTS Aan Autoregressive ahayn

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Waa maxay macnaha 'non-autoregressive' macnaha guud ee FastSpeech?
   • Waxay soo saartaa dhammaan xayndaabyada spectrogram si isbarbar socda halkii ay ka ahaan lahayd si isdaba joog ah
   • Waxa ay soo saartaa xayndaab maqal ah midba mar si isku xigta
   • Ma xamili karo weedho dhaadheer
   • Uma baahna xog tababar
2. Dhibaato noocee ah sida Tacotron 2 ayaa FastSpeech si gaar ah wax uga qabtaa?
   • Awood la'aanta soo saarista maqal kasta
   • Jiilka qunyar socodka ah iyo ereyada lagu soo celceliyay ee khaladaadka dareenka
   • Taageero la'aanta qoraalka Ingiriisiga
   • Baahida qalabka quantum
3. Waa maxay doorka 'xakameeyaha dhererka' ee FastSpeech?
   • Waxay cadaadisaa faylka maqalka ah ee ugu dambeeya
   • Waxay ka saartaa aamusnaanta duubista
   • Waxay ku soo celisaa matalida fone kasta iyadoo loo eegayo muddada la saadaaliyay si ay u waafaqdo dhererka maqalka
   • Waxay u tarjumeysaa teebabka qoraal
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. Waa maxay sababta FastSpeech's u fiirsigu si dirqi ah ugu koraan dhererka jumlada?
   • Dhammaan xirmooyinka waxaa loo soo saaray si isbarbar socda halkii ay ahaan lahaayeen kuwo isku xiga
   • Way iska indho tirtaa inta badan qoraalka la geliyo
   • Kaliya waxay soo saartaa hal-ilbiriqsi clips
   • Waxay adeegsataa shax raadin oo weedho hore loo duubay
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