WaveNet: A Generative Model for Raw Audio
2016 · paper · AI and machine learning
Introduces WaveNet, a neural network that generates raw audio waveforms sample by sample using dilated causal convolutions. In text-to-speech, listeners rated it as more natural than the best existing systems; it also generates music and shows promise for phoneme recognition.
Checked: PDF from arxiv.org, fetched 4 Oct 2026. Page numbers are those of that PDF.
0 findings
None that passed a second check.
9 minor (typos and the like; most checked by the judge only)
Wording "WaveNet outperformed the baseline statistical parametric and concatenative speech synthesizers in both languages" names no variant. In Table 2, Mandarin LSTM vs WaveNet (L) is 25.0 vs 23.3 (p = 0.476, not bold), so the claim holds only for WaveNet (L+F).
- p. 6 Section 3.2, paragraph on Fig. 5
WaveNet outperformed the baseline statisti-cal parametric and concatenative speech synthesizers in both languages
- p. 15 Table 2, Mandarin row LSTM vs WaveNet (L)
0.476
Judge's note: Checked the p.6 text and the Table 2 crop on p.15. In the Mandarin row LSTM vs WaveNet (L), neither side is bold and p = 0.476 > 0.01. Every other WaveNet-vs-baseline row favours WaveNet significantly. The sentence follows a reference to Fig. 5, whose bottom panel compares WaveNet (L+F) with the best baseline, so the authors may mean L+F. The next sentence, however, discusses WaveNet (L), so the scope is ambiguous. This is a wording point, not a data error.
Second check: another model agreed
Report a problem with this finding
Typo Subject–verb agreement: "The main ingredient of WaveNet are causal convolutions." should read "is".
- p. 2 Section 2.1, paragraph after Figure 2
The main ingredient of WaveNet are causal convolutions.
Judge's note: Seen on the p.2 crop, in the paragraph after Figure 2. The singular subject 'ingredient' takes the plural verb 'are'. This is cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Typo in §3.3: "For out third set of experiments" should read "For our third set of experiments".
- p. 6 Section 3.3 Music, first sentence
For out third set of experiments we trained WaveNets
- p. 6 Section 3.3 Music, first line
For out third set of experiments
Judge's note: Visible on the p.6 page image (C003-a crop), so it is not an extraction artefact.
Second check: none (judge only)
Report a problem with this finding
Typo Subject–verb agreement: "Table 1 show the MOS test results." should read "Table 1 shows".
- p. 6 Section 3.2, paragraph introducing Table 1
Table 1 show the MOS test results.
Judge's note: Visible on the p.6 crop. Appendix B uses the correct form, 'Table 2 shows'.
Second check: none (judge only)
Report a problem with this finding
Typo WaveNet (L+F) is described as conditioned on "log F0 values" in the Fig. 5 caption, but on "linguistic features and F0 values" in the Table 2 caption.
- p. 15 Table 2 caption
conditioned on both linguistic features and F0 values
- p. 7 Figure 5 caption
both linguistic features and log F0 values
Judge's note: Compared the two captions (p.7 and p.15). §3.2 (line 281) and Appendix B state that the conditioning uses log F0, so the Table 2 caption, like the §3.2 discussion text, uses 'F0' as shorthand. This is a minor terminology inconsistency; either form is understandable.
Second check: none (judge only)
Report a problem with this finding
Typo Reference title typo: "Vocaine the vocoder and applications is speech synthesis" probably should read "applications in speech synthesis".
- p. 9 References, Agiomyrgiannakis 2015
Vocaine the vocoder and applications is speech synthesis.
- p. 9 References, Agiomyrgiannakis 2015
Vocaine the vocoder and applications is speech synthesis
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the published title of Agiomyrgiannakis (ICASSP 2015) reads 'applications in speech synthesis'. The page image (p.9) shows 'is', and 'is' makes the title ungrammatical, so 'in' is almost certainly intended.
Second check: none (judge only)
Report a problem with this finding
Typo Malformed author list in the Fan et al. 2014 entry: "Fan, Yuchen, Qian, Yao, and Xie, Feng-Long, Soong Frank K." ('and' is misplaced and the comma after the surname 'Soong' is missing).
- p. 9 References, Fan et al. 2014
Fan, Yuchen, Qian, Yao, and Xie, Feng-Long, Soong Frank K.
- p. 9 References, Fan et al.
and Xie, Feng-Long, Soong Frank K.
- p. 11 References, Takamichi et al.
Sakriani, Sakti, and Naka-
Judge's note: Seen on the p.9 crop. Every other entry uses 'Surname, Given, ..., and Surname, Given'. This entry puts 'and' before the third author and gives the fourth as 'Soong Frank K.' with no comma. The Takamichi name-order point that one reviewer also raised is judged separately under C008.
Second check: none (judge only)
Report a problem with this finding
Typo In the Takamichi et al. 2016 entry, a co-author is printed as "Sakriani, Sakti" under the list's 'Surname, Given' convention. This would presumably be 'Sakti, Sakriani'.
- p. 11 References, Takamichi et al. 2016
Neubig, Graham, Sakriani, Sakti, and Naka-
- p. 11 References, Takamichi et al. 2016
Neubig, Graham, Sakriani, Sakti, and Naka-mura, Satoshi
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: Sakti is this author's family name and Sakriani the given name. The page image (p.11) shows 'Neubig, Graham, Sakriani, Sakti, and Nakamura, Satoshi'. Under the list's own convention, that order makes 'Sakriani' the surname.
Second check: none (judge only)
Report a problem with this finding
Typo Misspelling in Appendix A: "statistical parametric speech syntheziers matched" should read "synthesizers".
- p. 13 Appendix A, paragraph after Eq. (5)
statistical parametric speech syntheziers matched
- p. 13 Appendix A, paragraph after Eq. 5
speech syntheziers matched
- p. 13
state-of-the-art statistical parametric speech syntheziers matched state-of-the-art con-
Judge's note: Visible on the p.13 crop. Everywhere else the paper spells it 'synthesizers'.
Second check: none (judge only)