Visually grounded learning of keyword prediction from untranscribed speech
This work addresses the challenge of speech processing without transcriptions for robots or AI systems, but it is incremental as it builds on existing methods for mapping images and speech into a common space.
The paper tackled the problem of predicting keywords from untranscribed speech by using visual cues from paired images, resulting in a speech system that acts as a spoken bag-of-words classifier without parallel speech-text data, often confusing semantically related words like 'man' and 'person' to enhance semantic keyword spotting.
During language acquisition, infants have the benefit of visual cues to ground spoken language. Robots similarly have access to audio and visual sensors. Recent work has shown that images and spoken captions can be mapped into a meaningful common space, allowing images to be retrieved using speech and vice versa. In this setting of images paired with untranscribed spoken captions, we consider whether computer vision systems can be used to obtain textual labels for the speech. Concretely, we use an image-to-words multi-label visual classifier to tag images with soft textual labels, and then train a neural network to map from the speech to these soft targets. We show that the resulting speech system is able to predict which words occur in an utterance---acting as a spoken bag-of-words classifier---without seeing any parallel speech and text. We find that the model often confuses semantically related words, e.g. "man" and "person", making it even more effective as a semantic keyword spotter.