Whisper Timestamped Word Alignment Quiz

Answers and explanations appear as you take the quiz.

1. What does word-level alignment add that Whisper's native output lacks?
   • A start and end time for every individual word
   • Higher transcription accuracy
   • Speaker identity labels
   • Background noise removal
2. Which internal signal does whisper-timestamped use to locate words in time?
   • The audio waveform amplitude
   • The decoder's cross-attention weights over audio frames
   • The optimizer's gradients
   • The tokenizer vocabulary size
3. Why does whisper-timestamped use Dynamic Time Warping during alignment?
   • To translate the transcript into another language
   • To identify the speaker without using audio
   • To find an ordered, monotonic path between token and audio-frame scores
   • To guarantee exact boundaries for every word
4. How does WhisperX sharpen word boundaries compared to raw attention?
   • It increases the sampling rate
   • It uses a larger Whisper checkpoint only
   • It removes punctuation
   • It runs a separate phoneme-based forced-alignment model on the text
5. At what rate does Whisper's encoder produce audio frames?
   • About 50 frames per second (one every 20 ms)
   • Exactly 1 frame per second
   • 1000 frames per second
   • One frame per word
6. In what length chunks does Whisper process audio?
   • 5-second chunks
   • 30-second chunks
   • 2-minute chunks
   • Single samples
7. What does forced alignment assume is already available?
   • A speaker identity for every sound
   • A translation in every language
   • A transcript to align with the audio
   • A list of subtitle colors
8. What limitation should users keep in mind when interpreting word timestamps?
   • They prove the transcript is error-free
   • They always identify the speaker correctly
   • They are guaranteed to have millisecond accuracy
   • They are estimates and can be wrong when speech or alignment is difficult