Average Jaccard Similarity on Noisy and Reverberant Speech Utterances t60 = 0.7s
61Average Jaccard Similarity (-5dB)BEST-STD 2.0
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BEST-STD 2.0Type=Speech Tokens, Tokens=10242025.12 | 61 | 69 | 74 | 77 | 79 | |
| Ours - TransformerType=Speech Tokens, Tokens=10242025.12 | 57 | 64 | 68 | 72 | 73 | |
| WavLM-LargeType=ASR Posteriors, Tokens=322025.12 | 52 | 60 | 65 | 68 | 70 | |
| HuBERT-LargeType=ASR Posteriors, Tokens=322025.12 | 24 | 37 | 49 | 58 | 64 | |
| BEST-STDType=Speech Tokens, Tokens=10242025.12 | 19 | 22 | 38 | 55 | 62 | |
| WavLM-LargeType=Speech Tokens, Tokens=10002025.12 | 16 | 17 | 18 | 18 | 20 | |
| SpeechTokenizerType=Speech Tokens, Tokens=10242025.12 | 3 | 4 | 5 | 7 | 8 |