Segmenting multi-intent queries for spoken language understanding
Authors: Rohan Shet, Elena Davcheva, Christian Uhle
Abstract:
With the rising popularity of voice assistants, automatic speech recognition (ASR) systems play a crucial role in translating speech to text in order to enablenatural language understanding (NLU) models to process human commands. However, NLU models lack the proper resources to handle certain natural human speechcharacteristics, such as dividing or segmenting several intents expressed in one spo-ken sentence. This study presents an innovative approach to sentence boundarysegmentation from ASR output using a neural network model. We improve on previous attempts by removing the need for complex model output postprocessing, aswell as reporting higher accuracy than previous studies on the subject.