Language Modelling On Salmon
Metrics
Background (Domain) Consistency
Background (Random) Consistency
Background Alignment
Gender Consistency
Room Consistency
Sentiment Alignment
Sentiment Consistency
Speaker Consistency
Results
Performance results of various models on this benchmark
Model Name | Background (Domain) Consistency | Background (Random) Consistency | Background Alignment | Gender Consistency | Room Consistency | Sentiment Alignment | Sentiment Consistency | Speaker Consistency | Paper Title | Repository |
|---|---|---|---|---|---|---|---|---|---|---|
| Spirit-LM (Expr.) | 55.0 | 64.0 | 59.5 | 85.0 | 54.5 | 52.0 | 73.5 | 81.0 | Spirit LM: Interleaved Spoken and Written Language Model | - |
| LAST 350M | 55.5 | 60.5 | 54.5 | 70.5 | 61.0 | 51.5 | 64.0 | 63.0 | LAST: Language Model Aware Speech Tokenization | - |
| TWIST 1.3B | 55.5 | 60.5 | 56.5 | 69.5 | 59.0 | 53.0 | 61.5 | 69.0 | Textually Pretrained Speech Language Models | |
| TWIST 350M | 54.0 | 61.5 | 56.5 | 68.0 | 59.0 | 51.5 | 59.0 | 69.5 | Textually Pretrained Speech Language Models | |
| pGSLM | 57.0 | 66.0 | 53.5 | 88.5 | 53.5 | 55.5 | 40.5 | 83.0 | Text-Free Prosody-Aware Generative Spoken Language Modeling | |
| LAST 1.3B | 56.0 | 61.0 | 53.0 | 68.5 | 62.5 | 53.5 | 65.0 | 64.5 | LAST: Language Model Aware Speech Tokenization | - |
| TWIST 7B | 55.0 | 60.5 | 54.5 | 70.0 | 62.0 | 51.5 | 61.5 | 71.0 | Textually Pretrained Speech Language Models | |
| Spirit-LM (base) | 53.5 | 55.5 | 51.5 | 67.0 | 54.5 | 48.0 | 54.5 | 69.5 | Spirit LM: Interleaved Spoken and Written Language Model | - |
0 of 8 row(s) selected.