Maswali yanayounganishwa na mwongozo · Kati Kiwango

Maswali ya Utambuzi wa Hisia ya Hotuba

Hukagua viashiria vya sauti, mipaka ya seti ya data, ujanibishaji mtambuka, utata wa lebo, na marufuku finyu ya mahali pa kazi na elimu ya Umoja wa Ulaya.

Njia za mwongozo zinazohusianaUtambuzi wa Hisia za Hotuba
Swali 1 ya 8

Je, Utambuzi wa Hisia za Hotuba huchanganua nini kimsingi?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. Je, Utambuzi wa Hisia za Hotuba huchanganua nini kimsingi?
    • Vipengele vya akustisk ya sauti kama sauti na nishati
    • Maneno halisi katika nakala
    • Sarufi ya sentensi
    • Ishara za uso za mzungumzaji
  2. Which acoustic properties can speech-emotion models use as cues?
    • Only the spelling of transcribed words
    • Pitch, energy, speaking rate, and voice-quality features
    • A person’s job title
    • The number of speakers in a company
  3. Kwa nini kuweka lebo data ya hisia ni ngumu sana?
    • Faili za sauti ni kubwa mno kuhifadhi
    • Maikrofoni haiwezi kurekodi hisia
    • Hisia ni ya kibinafsi na wachambuzi wa kibinadamu mara nyingi hawakubaliani
    • Hisia hazionekani kamwe katika hotuba
  4. Ni ipi kati ya hizi ni mkusanyiko wa data wa usemi wa hisia unaojulikana sana?
    • Picha ya Picha
    • COCO
    • KIKOSI
    • IEMOCAP
  5. Je, mifumo ya kisasa ya SER mara nyingi hutumia vipi data iliyo na lebo?
    • Kwa kurekebisha vizuri miundo inayojisimamia kama wav2vec 2.0
    • Kwa kuzuia mitandao ya neva kabisa
    • Kwa kutumia sheria zilizoandikwa kwa mkono pekee
    • Kwa kurekodi mamilioni ya klipu mpya za hisia kwanza
  6. What does the EU AI Act prohibit regarding emotion recognition?
    • All emotion-related research in any setting
    • Inferring emotions in workplace or educational settings, except for medical or safety reasons
    • Every voice assistant that detects speech
    • Any system that transcribes spoken words
  7. Why can a speech-emotion model trained on one dataset fail on another?
    • Emotion labels are identical across every corpus
    • Microphone and speaker differences never affect audio
    • Speakers, language, recording conditions, elicitation, and label distributions can differ
    • A trained model cannot process new recordings
  8. What is one limitation of treating a speech-emotion prediction as a person’s true internal state?
    • Acoustic features contain no information
    • A predicted label is a clinical diagnosis by default
    • The same voice always expresses the same emotion
    • A model infers patterns from audio and may not reflect the speaker’s context or self-report