Semantic-Acoustic Alignment on SALMon
93.3Sentiment ScoreHuman
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HumanEvaluator=Human Topline2026.01 | 93.3 | 95.7 | |
| Flow-SLMConfig=270M, Evaluator=Global-PPL2026.01 | 59 | 53.5 | |
| Flow-SLMConfig=1B, Evaluator=Global-PPL2026.01 | 58.5 | 54.5 | |
| Flow-SLMConfig=1B, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 58 | 53.5 | |
| Flow-SLMConfig=1B-ext., Evaluator=Global-PPL2026.01 | 57.5 | 53.5 | |
| GSLMEvaluator=Windowed Likelihood-Based Evaluation2026.01 | 56 | 43 | |
| GSLMEvaluator=Global-PPL2026.01 | 55 | 52.5 | |
| Flow-SLMConfig=270M, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 54.5 | 55 | |
| TWISTConfig=1.3B, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 54 | 55 | |
| TASTEConfig=emb, Evaluator=Global-PPL2026.01 | 54 | 42 | |
| pGSLMEvaluator=Global-PPL2026.01 | 53 | 54 | |
| Spirit LMConfig=Expr., Evaluator=Global-PPL2026.01 | 53 | 59.5 | |
| Spirit LMConfig=Expr., Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 53 | 55 | |
| Llama-mimiConfig=1.3B, Evaluator=Global-PPL2026.01 | 53 | 49 | |
| TWISTConfig=1.3B, Evaluator=Global-PPL2026.01 | 52.5 | 57 | |
| Flow-SLMConfig=1B-ext., Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 52 | 57 | |
| TWISTConfig=350M, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 51.5 | 58 | |
| TWISTConfig=7B, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 51.5 | 56.5 | |
| TASTEConfig=emb, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 51.5 | 48 | |
| TWISTConfig=350M, Evaluator=Global-PPL2026.01 | 51 | 56.5 | |
| TWISTConfig=7B, Evaluator=Global-PPL2026.01 | 51 | 53.5 | |
| Llama-mimiConfig=1.3B, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 51 | 53.5 | |
| pGSLMEvaluator=Windowed Likelihood-Based Evaluation2026.01 | 49 | 52.5 | |
| Spirit LMConfig=base, Evaluator=Global-PPL2026.01 | 47.5 | 51 | |
| Spirit LMConfig=base, Evaluator=Windowed Likelihood-Based Evaluation2026.01 | 46 | 60.8 |