Speech-to-speech translation on CVSS-C DE → EN (test)
75.53ASR-BLEUGround Truth
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ground Truth2025.11 | 75.53 | — | |
| RosettaSpeech (Paralleled, Fine-tuned)Protocol=Non Zero-Shot, Training setup=Paralleled, Initialization=Fine-tuned, Model scope=Pair-specific models2025.11 | 29.9 | 32.62 | |
| RosettaSpeech (Unparalleled)Protocol=Zero-Shot, Training setup=Unparalleled2025.11 | 25.17 | 31.95 | |
| RosettaSpeech (Paralleled, Fine-tuned)Protocol=Non Zero-Shot, Training setup=Paralleled, Initialization=Fine-tuned, Model scope=Single model for all language pairs2025.11 | 21.54 | 23.22 | |
| StreamSpeechProtocol=Non Zero-Shot, Decoding strategy=Beam search (size 10)2025.11 | 20.93 | 23.36 | |
| StreamSpeechProtocol=Non Zero-Shot, Decoding strategy=Greedy search2025.11 | 19.72 | 21.96 | |
| UnitYProtocol=Non Zero-Shot, Decoding strategy=Beam search (size 10)2025.11 | 18.74 | — | |
| UnitYProtocol=Non Zero-Shot, Decoding strategy=Greedy search2025.11 | 18.19 | — | |
| ComSpeechProtocol=Zero-Shot2025.11 | 18.16 | 19.41 | |
| Translatotron 2Protocol=Non Zero-Shot, Decoding strategy=Beam search (size 10)2025.11 | 16.91 | — | |
| Translatotron 2Protocol=Non Zero-Shot, Decoding strategy=Greedy search2025.11 | 16.24 | 18.66 | |
| DASpeechProtocol=Non Zero-Shot2025.11 | 16.14 | — | |
| RosettaSpeech (Paralleled, Scratch)Protocol=Non Zero-Shot, Training setup=Paralleled, Initialization=Scratch2025.11 | 9.02 | 17.71 | |
| S2UTProtocol=Non Zero-Shot, Decoding strategy=Beam search (size 10)2025.11 | 2.99 | — | |
| S2UTProtocol=Non Zero-Shot, Decoding strategy=Greedy search2025.11 | 2.46 | — | |
| TranslatotronProtocol=Non Zero-Shot2025.11 | 1.97 | — | |
| Qwen3-0.6B2025.11 | — | 27.42 |