Ni afiwe WaveGAN Vocoder adanwo

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

1. Kini iṣẹ ti vocoder bi Parallel WaveGAN?
   • Yipada mel-spectrogram kan si ọna igbi ohun aise
   • Yi ọrọ pada sinu awọn foonu foonu
   • Yọ ariwo abẹlẹ kuro lati awọn igbasilẹ
   • Tumọ ọrọ laarin awọn ede
2. Bawo ni Parallel WaveGAN ṣe ipilẹṣẹ awọn ayẹwo ohun ni akawe pẹlu atilẹba WaveNet?
   • Apeere kan ni akoko kan, lẹsẹsẹ
   • Gbogbo awọn ayẹwo ni afiwe, ti kii-autoregressively
   • Nipa gbigbasilẹ gbohungbohun
   • Nipa sisopọ awọn agekuru ti o fipamọ
3. Ipadanu wo ni aringbungbun si Parallel WaveGAN lẹgbẹẹ ipadanu ọta?
   • Cross-entropy lori awọn ami ọrọ
   • Triplet ipo pipadanu
   • Olona-o ga STFT pipadanu
   • Ẹsan imuduro
4. Ohun faaji wo ni Parallel WaveGAN monomono lo?
   • LSTM loorekoore
   • A iran transformer
   • Igi ipinnu
   • Nẹtiwọọki ara WaveNet kan ti o fẹ
5. Kini idi ti Parallel WaveGAN wuni fun imuṣiṣẹ?
   • O ti wa ni iwapọ ati ki o nṣiṣẹ yiyara ju akoko gidi
   • O jẹ awoṣe paramita pupọ-pupọ pupọ pupọ
   • Ko nilo igbewọle spectrogram
   • O ṣiṣẹ nikan lori supercomputers
6. Ninu GAN kan, ipa wo ni ẹlẹyamẹya ṣe lakoko ikẹkọ?
   • O ṣe ipilẹṣẹ spectrogram
   • O ṣe idajọ boya ohun ohun dabi gidi tabi iro
   • O ṣe igbasilẹ iwe-ipamọ data
   • O tọju awọn iwuwo awoṣe
7. In Parallel WaveGAN, what does the noise input contribute alongside mel-spectrogram conditioning?
   • The speaker transcript after synthesis
   • A replacement for the discriminator
   • Stochastic input for waveform generation while the mel features condition the acoustic content
   • A fixed random waveform copied into every output
8. What does the adversarial loss encourage the Parallel WaveGAN generator to produce?
   • Text transcripts with perfect spelling
   • Audio samples with no speaker conditioning
   • A smaller mel-spectrogram vocabulary
   • Waveforms that the discriminator judges as resembling real speech