Ikibazo gihuza ibibazo · Hagati Urwego
Ikibazo cya RNN-Transducer
Sobanukirwa uburyo moderi ya RNN-Transducer ihuza amakadiri ya acoustic namateka yo gusohora kugirango akurikirane imirimo yo kwandukura.
Ikibazo 1 Bya 8
Ni izihe mbogamizi za CTC RNN-Transducer ikemura?
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- Ni izihe mbogamizi za CTC RNN-Transducer ikemura?
- CTC isabwa-kwigenga kwifata, irinda kwerekana ibyasohotse-token biterwa
- CTC ntishobora gutunganya amajwi na gato
- CTC isaba GPU gukora
- CTC ikora amashusho gusa
- Nibihe bintu bitatu byingenzi bigize RNN-Transducer?
- Encoder, decoder, nivangura
- Encoder (umuyoboro wandikirwa), umuyoboro wo guhanura, hamwe numuyoboro uhuriweho
- Tokenizer, gushiramo, na softmax gusa
- Imitwe ibiri yitonderwa hamwe na router
- Ni uruhe ruhare umuyoboro uhanura ugira muri RNN-T?
- Ihagarika amajwi yerekana amajwi
- Ipima ingano ya mikoro
- Ikora nkicyitegererezo cyururimi, itondekanya ibimenyetso byasohotse mbere
- Ihitamo igipimo cyicyitegererezo
- Kuki RNN-T ishyigikira umuvuduko muke gutemba muburyo busanzwe?
- Itunganya imvugo yose icyarimwe mbere yo kohereza ikintu icyo aricyo cyose
- Irengagiza ibihe byose
- Ikora gusa kuri dosiye zanditswe mbere
- Guhuza kwayo ni monotonic n'ibumoso-iburyo, bitera imbere binyuze mumajwi hamwe nubusa
- Muri decode ya RNN-T, gusohora ikimenyetso cyambaye iki?
- Iterambere kumwanya ukurikira wamajwi
- Kurangiza inyandiko yose
- Gusiba ijambo ryabanje
- Hindura urundi rurimi
- Ni ubuhe bwoko bwa encoder bugenda busimbuza LSTM muri sisitemu igezweho ya sisitemu ya RNN-T?
- Ibiti
- Kode ya kodegisi
- Umufuka wuzuye-w-amagambo
- Icyitegererezo cya Markov
- What two histories condition the RNN-T joint network’s output distribution?
- Only the future transcript and a speaker photo
- Only the last acoustic frame, without output context
- The encoder representation of acoustic input and the prediction-network representation of prior output labels
- The optimizer state and learning rate
- What alignment behavior does an RNN-T allow before advancing to the next audio frame?
- It must emit exactly one label for every frame
- It advances time only after a word-final token
- It generates all output before reading any audio
- It can emit zero or more nonblank labels at the current frame, then a blank advances time