Metric: Sentiment Consistency (higher is better)
| # | Model↕ | Sentiment Consistency▼ | Extra Data | Paper | Date↕ | Code |
|---|---|---|---|---|---|---|
| 1 | Spirit-LM (Expr.) | 73.5 | Yes | Spirit LM: Interleaved Spoken and Written Langua... | 2024-02-08 | Code |
| 2 | LAST 1.3B | 65 | Yes | LAST: Language Model Aware Speech Tokenization | 2024-09-05 | - |
| 3 | LAST 350M | 64 | Yes | LAST: Language Model Aware Speech Tokenization | 2024-09-05 | - |
| 4 | TWIST 7B | 61.5 | Yes | Textually Pretrained Speech Language Models | 2023-05-22 | Code |
| 5 | TWIST 1.3B | 61.5 | Yes | Textually Pretrained Speech Language Models | 2023-05-22 | Code |
| 6 | TWIST 350M | 59 | Yes | Textually Pretrained Speech Language Models | 2023-05-22 | Code |
| 7 | TASLM 1B (token) | 59 | No | - | - | - |
| 8 | TASLM 1B (embedding) | 57.5 | No | - | - | - |
| 9 | Spirit-LM (base) | 54.5 | Yes | Spirit LM: Interleaved Spoken and Written Langua... | 2024-02-08 | Code |
| 10 | pGSLM | 40.5 | Yes | Text-Free Prosody-Aware Generative Spoken Langua... | 2021-09-07 | Code |