Post by account_disabled on Mar 11, 2024 2:05:35 GMT -7
Facebook AI features a non-text-based NLP model. It's the first step towards a new approach to natural language processing that appears to have huge benefits. Alessio Pomaro Alessio Pomaro Sep 23, 2021 •5 min read Facebook AI introduces GSLM: a non-text-based NLP model Facebook AI introduces GSLM: a non-text-based NLP model Text-based language models like BERT , RoBERTa , and GPT-3 have made great strides in recent years. By providing written sentences as input, they can generate highly realistic text on virtually any topic. Furthermore, they also give rise to useful pre-trained models that can be made available to a wide variety of complex natural language processing ( NLP ) applications, such as sentiment analysis , translation systems , entity and information recognition, inferences and summaries, using a few labels and examples.
These applications, however, highlight an important limitation: they can India Mobile Number Data only be used effectively on languages equipped with large textual datasets , capable of training AI models. Generative Spoken Language Model (GSLM) What is GSLM? It is a Facebook AI project , and stands for Generative Spoken Language Model , which is the first NLP model that frees itself from dependence on text. A model that opens the doors to a new era of NLP applications potentially capable of being used for all languages spoken on Earth, even those without large textual datasets. GSLM also enables the development of NLP models that incorporate the full range of expressiveness of oral language .
Until now, connecting an NLP application to vocal inputs involved training an automatic speech recognition system ( Automatic Speech Recognition - ASR ), an operation characterized by a high effort and the introduction of potential errors, with less than optimal coding performance, and available for a limited number of languages. With these new models, the goal is to make ASR obsolete by going " end-to-end " from voice input to output. The following video shows how AI, thanks to GSLM, manages to continue a vocal input from a person reading a fragment of a well-known book. An example application of GSLM where the AI continues a voice input in 3 versions It is impressive that the speech generated by the model is absolutely aligned, and a listener would hardly perceive the change.
These applications, however, highlight an important limitation: they can India Mobile Number Data only be used effectively on languages equipped with large textual datasets , capable of training AI models. Generative Spoken Language Model (GSLM) What is GSLM? It is a Facebook AI project , and stands for Generative Spoken Language Model , which is the first NLP model that frees itself from dependence on text. A model that opens the doors to a new era of NLP applications potentially capable of being used for all languages spoken on Earth, even those without large textual datasets. GSLM also enables the development of NLP models that incorporate the full range of expressiveness of oral language .
Until now, connecting an NLP application to vocal inputs involved training an automatic speech recognition system ( Automatic Speech Recognition - ASR ), an operation characterized by a high effort and the introduction of potential errors, with less than optimal coding performance, and available for a limited number of languages. With these new models, the goal is to make ASR obsolete by going " end-to-end " from voice input to output. The following video shows how AI, thanks to GSLM, manages to continue a vocal input from a person reading a fragment of a well-known book. An example application of GSLM where the AI continues a voice input in 3 versions It is impressive that the speech generated by the model is absolutely aligned, and a listener would hardly perceive the change.