Automatic Speech Recognition on Multilingual ASR Corpus Turkish - tr (test)
10.8CERembedding-based H-Softmax
Evaluation results
| Method | Links | |
|---|---|---|
| embedding-based H-SoftmaxTraining=All, Embedding Model=LABSE, Distance Metric=Median Euc2025.01 | 10.8 | |
| embedding-based H-SoftmaxTraining=All, Strategy=Mono-Map, Weighting=Avg CB2025.01 | 10.8 | |
| embedding-based H-SoftmaxTraining=All, Embedding Model=XLM-base, Distance Metric=2-med S-Euc2025.01 | 10.9 | |
| H-Softmax (Huffman)Training=All, Softmax Type=Huffman2025.01 | 11 | |
| embedding-based H-SoftmaxTraining=All, Embedding Model=XLM-large, Distance Metric=2-med S-Euc2025.01 | 11.2 | |
| embedding-based H-SoftmaxTraining=All, Embedding Model=XLM-XL, Distance Metric=Wtd Corr2025.01 | 11.7 | |
| embedding-based H-SoftmaxEmbedding Source=Mono-Map, Clustering Method=Average, Distance Metric=City Block2025.01 | 12.5 | |
| embedding-based H-SoftmaxEmbedding Source=XLM-base, Clustering Method=2-medoids, Distance Metric=Standardized Euclidean2025.01 | 12.6 | |
| embedding-based H-SoftmaxEmbedding Source=XLM-large, Clustering Method=2-medoids, Distance Metric=Standardized Euclidean2025.01 | 12.7 | |
| Huffman H-SoftmaxOutput Strategy=H-Softmax, Clustering=Huffman2025.01 | 12.8 | |
| embedding-based H-SoftmaxEmbedding Source=XLM-XL, Clustering Method=Weighted, Distance Metric=Correlation2025.01 | 13.2 | |
| embedding-based H-SoftmaxEmbedding Source=LABSE, Clustering Method=Median, Distance Metric=Euclidean2025.01 | 13.2 | |
| SoftmaxOutput Strategy=Softmax2025.01 | 13.6 | |
| SoftmaxTraining=All, Softmax Type=Standard Softmax2025.01 | 13.7 |