medium10 pts · first answer countstransformersmultimodaltransformer-architecture

How Whisper-style speech models work

Whisper-style speech-to-text models are transformers. How are they usually structured?

A.An encoder reads the audio spectrogram; a decoder writes text tokensB.A decoder-only model reads raw audio bytes as text tokensC.An encoder-only model outputs one label for the whole clipD.Two separate decoders, one per language, vote on each word

Discussion 0

Join free to post in the discussion.

How Whisper-style speech models work

Discussion 0

Related questions