kahne/SpeechTransProgress

Comprehensive resource compiling multilingual speech translation datasets (CoVoST 2, CVSS, mTEDx, MUST-C) spanning 20+ language pairs with both text and speech targets, alongside implementations in major frameworks like ESPNet-ST and Fairseq S2T. Tracks benchmark progress through peer-reviewed papers and tutorials covering direct speech-to-text translation without intermediate ASR, end-to-end architectures that jointly model acoustic and linguistic knowledge. Provides curated bibliography of recent advances including large language model fine-tuning approaches and comparative evaluations across diverse domain corpora ranging from TED talks to parliamentary proceedings.

261 stars. No commits in the last 6 months.