vvolhejn
commented
2 years ago - DDSP: no survey
- NEWT Sec 5.3: MUSHRA-like, 19 participants, statistical analysis
- MIDI-DDSP: Likert scale, 960 pairwise comparisons, statistical analysis.
- RAVE Sec 5.1: MOS, 33 participants, 15 trials per participant, 95% CI
- Tacotron 2 Sec 3.2: MOS and pairwise against GT, 100 examples, at least 8 ratings for each
- HTP Sec 4.3: MOS for timbre transfer, #people=?
- A Universal Music Translation Network Sec 4: MOS +- SD, timbre transfer questions: "what is the quality of the audio, and how well does the converted version match the origin", #people=?
- Unsupervised Singing Voice Conversion Sec 4: MOS for naturalness and similarity
- Transferring Neural Speech...: MOS for quality and similarity (neural vocoders), 150 samples * 9 systems, each evaluated 6 times, 136 people, statistical analysis
- Unsupervised Cross-Domain Singing Voice Conversion Sec 4: MOS for naturalness, 40 samples, 20 raters, STD?
Neural audio synthesis models are hard to compare automatically - a survey will be needed to show that the quality didn't decrease through our speedups