Speech-to-Speech Translation Quiz

Answers and explanations appear as you take the quiz.

1. What does Speech-to-Speech Translation (S2ST) do?
   • Converts speech in one language into spoken audio in another language
   • Only transcribes audio into text
   • Translates written documents to PDFs
   • Generates background music
2. What are the three stages of a classic cascaded S2ST system?
   • Record, store, delete
   • Speech recognition, machine translation, text-to-speech
   • Encrypt, compress, send
   • Listen, ignore, repeat
3. What is a key advantage of direct (end-to-end) S2ST over cascaded systems?
   • It always needs more text steps
   • It cannot translate at all
   • It can lower latency and better preserve voice and prosody
   • It only works for one language pair
4. Which Meta system is known for translating across roughly 100 languages?
   • ChatGPT
   • Midjourney
   • Stable Diffusion
   • SeamlessM4T
5. Why is real-time streaming translation especially challenging?
   • The system must decide when to translate before a sentence finishes
   • Audio cannot be streamed
   • Languages all share identical word order
   • It requires no models
6. What emerging concern comes from preserving a speaker's voice in translation?
   • Files become too small
   • Consent and deepfake/voice-cloning misuse
   • Translation becomes impossible
   • Audio loses all sound
7. What representation may a direct speech-to-speech system use to connect source audio to target audio?
   • A Slurm job identifier
   • A table of image pixels
   • Discrete acoustic units that serve as intermediate speech tokens
   • A hand-written class label for every word
8. Why can errors accumulate in a cascaded ASR → translation → TTS pipeline?
   • Each stage uses the same model weights
   • TTS always corrects the transcript
   • The stages run in parallel on the same audio
   • An upstream transcription or translation error becomes input to later stages