Speech-to-speech translation for a real-world unwritten language

Open paper
Benchmark Model Rank Results
speech-to-speech-translation-on-tatHokkien→En (Two-pass decoding)#1ASR-BLEU (Dev): 13.6ASR-BLEU (Test): 12.5
speech-to-speech-translation-on-tatHokkien→En (Two-stage)#2ASR-BLEU (Dev): 12.5ASR-BLEU (Test): 10.5
speech-to-speech-translation-on-tatHokkien→En (Three-stage)#3ASR-BLEU (Dev): 12.5ASR-BLEU (Test): 8.8
speech-to-speech-translation-on-tatHokkien→En (Single-pass decoding)#4ASR-BLEU (Dev): 8.8ASR-BLEU (Test): 8.1
speech-to-speech-translation-on-tatEn→Hokkien (Two-pass decoding)#5ASR-BLEU (Dev): 7.8ASR-BLEU (Test): 7.3
speech-to-speech-translation-on-tatEn→Hokkien (Three-stage)#6ASR-BLEU (Dev): 7.5ASR-BLEU (Test): 6.8
speech-to-speech-translation-on-tatEn→Hokkien (Two-stage)#7ASR-BLEU (Dev): 7.1ASR-BLEU (Test): 6.6
speech-to-speech-translation-on-tatEn→Hokkien (Single-pass decoding)#8ASR-BLEU (Dev): 6.6ASR-BLEU (Test): 6.0

Rank counts only results with a code link.