AI Learns to Speak Yorùbá With Just Six Hours of Audio
Researchers fine-tuned an English-trained FastSpeech 2 model with explicit phoneme-tone encoding to synthesize tonally accurate Yorùbá speech from just six ...
Researchers fine-tuned an English-trained FastSpeech 2 model with explicit phoneme-tone encoding to synthesize tonally accurate Yorùbá speech from just six ...
A controlled adaptation campaign for Modern Standard Arabic speech synthesis finds that synthetic diacritization, punctuation-guided length rescue, and longer training ...
A new study shows that synthetic speech generated by large language models significantly improves speech recognition for low-resource Estonian and ...
Kazakh researchers fine-tuned a multilingual sentence transformer with a staged contrastive objective on a trilingual paraphrase corpus, improving skill-concept matching ...
Researchers at Xizang University fine-tuned the DeepSeek-R1 reasoning model into DeepSeek-R1-Tibet, which outperformed CINO, mBERT, and TisBERT at inferring missing ...
Researchers fine-tuned a multilingual neural TTS model on 30.5 hours of Southern Quechua speech and achieved real-time synthesis directly on ...
A new review finds that Bangla medical question answering lags far behind English systems due to dataset scarcity, translated corpora, ...
A new framework combining policy gradient adversarial networks with MMD-guided sample transfer lifts machine translation BLEU scores to 39.5 and ...
Researchers have built a hierarchical attention network that detects fake news in English, Hindi, and Marathi, reaching 98.9 percent accuracy ...
Researchers in Pune have developed a pose-enhanced generative adversarial network that translates Marathi text into fluent Indian Sign Language gloss ...
Success! An email was just sent to confirm your subscription. Please find the email now and click 'Confirm Follow' to start subscribing.
© 2025 Scienmag - Science Magazine
© 2025 Scienmag - Science Magazine