Acoustic Evaluation on SALMon
79Consistency - SentimentLlama-Mimi-1.3B
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Llama-Mimi-1.3BTrain Data Size (Hours)=240K, Tokenizer=RVQ2025.09 | 79 | 85 | 83.5 | 73.5 | 92 | 48.5 | 53.5 | |
| CSM-1.3BTrain Data Size (Hours)=240K, Tokenizer=RVQ2025.09 | 73.5 | 77.5 | 75 | 70.5 | 81.5 | 45.5 | 54 | |
| SpiritLM-exp-7BTrain Data Size (Hours)=570K, Tokenizer=SSL2025.09 | 72 | 80.5 | 85.5 | 64 | 55.5 | 52.5 | 59.5 | |
| Flow-SLM-1B-extTrain Data Size (Hours)=110K, Tokenizer=RVQ2025.09 | 62.5 | 76 | 80 | 69 | 73 | 55 | 53 | |
| TWIST-1.3BTrain Data Size (Hours)=150K, Tokenizer=SSL2025.09 | 61.5 | 69 | 69.5 | 60.5 | 59 | 53 | 56.5 | |
| SpiritLM-base-7BTrain Data Size (Hours)=570K, Tokenizer=SSL2025.09 | 54 | 68 | 67 | 56 | 54.5 | 46 | 50.5 |