We believe that AI can and should benefit everyone. Increasing access to models like Llama 2 can help create a new era of economic and social opportunities for startups, entrepreneurs and researchers.
@aiatmeta
-
CM3leon: State-of-the-Art Multimodal Text-to-Image Generation Model
By
–
Introducing CM3leon, a first-of-its-kind multimodal model that achieves state-of-the-art performance for text-to-image generation with 5x the compute efficiency of competitive models. More details https://
bit.ly/44I6t7E -
Meta Open-Sources Speech Recognition Fairness Dataset
By
–
In line with this work, we're open-sourcing a new dataset to help the broader community improve fairness of speech recognition models. The dataset includes ~27K utterances in recorded speech from 595 paid participants. Dataset https://
bit.ly/3DiV0zr 3/3 -
Model Performance Improvements Across Demographics Without Fairness Tradeoffs
By
–
Early experimental results show that our approach improves model performance on all demographic groups in our evaluation datasets, with the largest gains in respect to accent inclusivity — demonstrating that improved fairness doesn’t mean a performance tradeoff. 2/3
-
Privacy-Preserving Approach Improves Speech Recognition Fairness
By
–
Today we're announcing a new privacy-preserving approach to improve fairness & robustness of automatic speech recognition systems. This unique approach lets researchers improve ASR performance without relying on demographic data. More info
-
Area Chair Award: Machine Translation with Minimal Data for Linguistic Diversity
By
–
Area Chair Award on Linguistic Diversity
Small Data, Big Impact: Leveraging Minimal Data for Effective Machine Translation https://
bit.ly/3POTFbl -
Language Model Acceptability Judgements Lack Contextual Robustness
By
–
ACL Outstanding Paper
Language model acceptability judgements are not always robust to context -
ACL Outstanding Paper on Hybrid Transducer Attention Encoder-Decoder Speech
By
–
ACL Outstanding Paper
Hybrid Transducer and Attention based Encoder-Decoder Modeling for Speech-to-Text Tasks -
Voicebox: Multilingual Speech Synthesis via Flow Matching
By
–
Voicebox can synthesize speech across six languages, perform noise removal, edit content, transfer audio style & more. It's based on Flow Matching, a new method proposed by Meta AI.
— AI at Meta (@AIatMeta) 6 juillet 2023
More details on this work & real examples ➡️ https://t.co/ptAtSE4DuQ pic.twitter.com/SUYbXw15oWVoicebox can synthesize speech across six languages, perform noise removal, edit content, transfer audio style & more. It's based on Flow Matching, a new method proposed by Meta AI. More details on this work & real examples
https://bit.ly/3CwUw8z -
Training Code and Model Checkpoints Released on GitHub
By
–
We've made the training code and model checkpoints available on GitHub and you can access them here https://
bit.ly/3NpVlWY