Speech Model Pre-training For End-to-end Spoken Language Understanding | Awesome LLM Papers Add your paper to Awesome LLM Papers

Awesome LLM Papers

A collection of research papers on LLMs.

Stay Updated

Contact Sean Moran about this survey or website. Made with Jekyll and Hyde.

Speech Model Pre-training For End-to-end Spoken Language Understanding

Loren Lugosch, Mirco Ravanelli, Patrick Ignoto, Vikrant Singh Tomar, Yoshua Bengio . Interspeech 2019 2019 – 245 citations

[Paper]   Search on Google Scholar   Search on Semantic Scholar
Compositional Generalization Datasets Interspeech Training Techniques

Whereas conventional spoken language understanding (SLU) systems map speech to text, and then text to intent, end-to-end SLU systems map speech directly to intent through a single trainable model. Achieving high accuracy with these end-to-end models without a large amount of training data is difficult. We propose a method to reduce the data requirements of end-to-end SLU in which the model is first pre-trained to predict words and phonemes, thus learning good features for SLU. We introduce a new SLU dataset, Fluent Speech Commands, and show that our method improves performance both when the full dataset is used for training and when only a small subset is used. We also describe preliminary experiments to gauge the model’s ability to generalize to new phrases not heard during training.

Similar Work