Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Idanwo Isọsọ oriṣi Orin

Covers audio representations, GTZAN’s composition, genre ambiguity, dataset leakage risks, and evaluation on artist-disjoint data.

Jẹmọ itọsọna awọn ọnaIsọri oriṣi Orin
Ibeere 1 ti 8

Kini mel-spectrogram, ti a lo nigbagbogbo bi titẹ sii si oriṣi-ipinsi awọn nẹtiwọọki nkankikan?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Kini mel-spectrogram, ti a lo nigbagbogbo bi titẹ sii si oriṣi-ipinsi awọn nẹtiwọọki nkankikan?
    • Aworan igba-akoko ti ohun nipa lilo iwọn ipolowo oye
    • Atokọ awọn orin orin ti yipada si awọn nọmba
    • Fọọmu igbi ti nfihan titobi aise lori akoko nikan
    • Data data ti awọn orukọ olorin ati awọn ọdun idasilẹ
  2. Iru data ala-ilẹ alailẹgbẹ wo ni o ni awọn agekuru iṣẹju-aaya 1,000 laarin awọn oriṣi 10?
    • Pipa Net
    • GTZAN
    • LibriSpeech
    • COCO
  3. Kini idi ti iyọrisi pipe-ipejuwe-isọtọ-pipe jẹ pataki pataki?
    • Awọn kọmputa ko le ṣe ilana iwe ohun rara
    • Gbogbo awọn orin dun aami si awọn algoridimu
    • Awọn aala oriṣi jẹ iruju ati ti aṣa
    • Awọn faili ohun ti kere ju lati ṣe itupalẹ
  4. Ẹya ti a ṣe ni ọwọ wo ni a lo ni lilo pupọ ni awọn ọna ṣiṣe iyasọtọ oriṣi akọkọ?
    • ratio funmorawon JPEG
    • HTML tag igbohunsafẹfẹ
    • Awọn histogram imọlẹ Pixel
    • Àwọn olùsọdipúpọ̀ cepstral-igbohunsafẹfẹ (MFCC)
  5. Kí ni a softmax Layer ni opin ti a oriṣi classifier gbejade?
    • Pipin iṣeeṣe lori awọn kilasi oriṣi
    • A nikan iwe igbi fọọmu
    • A fisinuirindigbindigbin MP3 faili
    • Aworan spectrogram kan
  6. How can self-supervised music pretraining reduce dependence on labeled genre examples?
    • It eliminates the need to evaluate genre labels
    • It can learn audio representations from unlabeled music, which are then evaluated or fine-tuned with labeled examples
    • It guarantees correct labels for every song
    • It converts genre categories into universal facts
  7. How can exact or repeated excerpts in GTZAN distort a random train/test evaluation?
    • They ensure that every genre label is objective
    • They make the test set contain more genres than training
    • If repetitions cross the split, a model may see duplicate or near-duplicate material in both sets
    • They remove any need to inspect class labels
  8. How do MFCCs differ from a mel-spectrogram representation?
    • MFCCs are the model’s final genre probabilities
    • A mel-spectrogram removes the time axis entirely
    • MFCCs measure only the recording’s file size
    • MFCCs summarize the log-mel spectrum with cepstral coefficients