गाइड से जुड़ी प्रश्नोत्तरी · कठिन स्तर

स्पीच रिकग्निशन क्विज़ में GMM-HMM ध्वनिक मॉडल

Test HMM hidden states and transitions, GMM acoustic emissions, lexicons, Viterbi decoding, model limits, and neural comparisons.

संबंधित मार्गदर्शक पथGmm Hmm Acoustic Models
प्रश्न 1 का 8

In a GMM-HMM recognizer, what does an HMM state commonly represent?

प्रत्येक प्रश्न का पूर्वावलोकन करें8 प्रश्न

जैसे ही आप प्रश्नोत्तरी लेते हैं उत्तर और स्पष्टीकरण दिखाई देते हैं।

  1. In a GMM-HMM recognizer, what does an HMM state commonly represent?
    • An exact audio-file checksum
    • A hidden phonetic or related speech state in a time sequence
    • The speaker’s permanent identity
    • A directly observed written letter
  2. What does the GMM component contribute to a candidate HMM state?
    • A guaranteed word spelling
    • A new acoustic recording
    • A language-independent identity certificate
    • A likelihood for the observed acoustic feature vector
  3. Why are state transitions useful in speech recognition?
    • Speech sounds occur in structured sequences over time
    • Every frame can be decoded independently
    • The lexicon becomes unnecessary
    • Word boundaries are directly observed
  4. How does a pronunciation lexicon help the conventional pipeline?
    • It measures emotion from audio
    • It replaces acoustic evidence with spelling
    • It connects candidate words with sound-state sequences
    • It guarantees every rare name is covered
  5. What does Viterbi-style decoding search for?
    • The certainly correct sentence
    • A high-probability hidden-state path through the sequence
    • A new image region
    • A microphone hardware setting
  6. Which assumption can limit a simple HMM?
    • The next state depends on limited previous-state context
    • All speakers have identical acoustics
    • Mixture components cannot have weights
    • Words do not require pronunciation models
  7. Why test on speakers and rooms absent from training?
    • A lexicon predicts every room condition
    • Training loss guarantees transfer
    • All speakers produce the same features
    • Clean in-source scores may not transfer across acoustics
  8. What distinguishes a GMM emission model from HMM transitions?
    • Both factors only count written words
    • Transitions score microphone brightness
    • Emissions score observed features under states; transitions score state changes
    • Emissions decide a school timetable