Metne Konuşma Sınavı

Testi çözdüğünüzde cevaplar ve açıklamalar görünür.

1. A transcript has four substitutions, three deletions, and two insertions against a 100-word reference. What is its word error rate?
   • 9%
   • 4%
   • 7%
   • %100
2. Why inspect individual transcript errors after calculating word error rate?
   • The score identifies speaker identity
   • The aggregate does not show whether an error changed a key instruction
   • The percentage verifies punctuation
   • The number proves accent fairness
3. A recording contains overlapping speakers and unusual names. What does the guide recommend?
   • Treat speaker labels as verified identity
   • Assume one dataset score applies to the recording
   • Expect recognition challenges and review names and uncertain passages
   • Remove the original audio
4. A transcript labels two speakers. What does a speaker label establish?
   • It shows that translation is correct
   • It proves who spoke
   • It guarantees transcription accuracy
   • It is a model output, not necessarily a verified identity
5. Why preserve the original recording and timestamps where permitted?
   • They automatically correct word error rate
   • They let reviewers check uncertain transcript passages against the source
   • They replace the transcript reference
   • They guarantee that no words were omitted
6. Which error types does word error rate count?
   • Speaker changes and audio duration
   • Only punctuation and capitalization
   • Substitutions, deletions, and insertions
   • Translation and summarization errors
7. Why report normalization rules with a WER score?
   • Punctuation, casing, and tokenization choices affect the result
   • Normalization verifies who spoke
   • Rules remove all semantic errors
   • It makes all datasets comparable automatically
8. A system transcribes silence as confident words. What does the guide suggest testing?
   • Speaker identity accuracy
   • Only clean speech from one speaker
   • Translation quality
   • Silence and non-speech audio