Spoken question answering for visual queries
This work addresses a dataset gap for multi-modal AI systems combining speech and images, but it is incremental as it builds on existing VQA and SQA methods.
The paper tackled the lack of a dataset for spoken visual question answering (SVQA) by synthesizing speech from text using zero-shot TTS models, and found that a model trained on synthesized speech nearly matches the performance of an upper-bound model trained on text, with TTS choice having minor impact on accuracy.
Question answering (QA) systems are designed to answer natural language questions. Visual QA (VQA) and Spoken QA (SQA) systems extend the textual QA system to accept visual and spoken input respectively. This work aims to create a system that enables user interaction through both speech and images. That is achieved through the fusion of text, speech, and image modalities to tackle the task of spoken VQA (SVQA). The resulting multi-modal model has textual, visual, and spoken inputs and can answer spoken questions on images. Training and evaluating SVQA models requires a dataset for all three modalities, but no such dataset currently exists. We address this problem by synthesizing VQA datasets using two zero-shot TTS models. Our initial findings indicate that a model trained only with synthesized speech nearly reaches the performance of the upper-bounding model trained on textual QAs. In addition, we show that the choice of the TTS model has a minor impact on accuracy.