Classification on VoxCeleb 1
72AccuracyF3-Tokenizer
Evaluation Results
| Method | Links | |
|---|---|---|
| F3-TokenizerProbing=Frozen representation2026.06 | 72 | |
| w/o LLMProbing=Frozen representation2026.06 | 65.4 | |
| w/o RQProbing=Frozen representation2026.06 | 56.8 | |
| WhisperProbing=Frozen representation2026.06 | 38.8 | |
| Ming-UProbing=Frozen representation2026.06 | 35.28 | |
| MetaMAEPretrain data=LibriSpeech, Protocol=Linear evaluation2023.10 | 7.4 | |
| No repr.Probing=Frozen representation2026.06 | 3.2 | |
| ShEDPretrain data=LibriSpeech, Protocol=Linear evaluation2023.10 | 2.8 | |
| e-MixPretrain data=LibriSpeech, Protocol=Linear evaluation2023.10 | 2.4 | |
| CapriPretrain data=LibriSpeech, Protocol=Linear evaluation2023.10 | 1.6 | |
| MAEPretrain data=LibriSpeech, Protocol=Linear evaluation2023.10 | 1.6 | |
| BaselinePretrain data=LibriSpeech, Protocol=Linear evaluation2023.10 | 0.6 |