Vision as an Interlingua: Learning Multilingual Semantic Embeddings of Untranscribed Speech
This addresses the challenge of processing speech in low-resource languages without transcriptions, though it is incremental as it extends prior monolingual work to multiple languages.
The paper tackles the problem of learning multilingual semantic embeddings from untranscribed speech and images, showing that a model trained on English and Hindi speech captions improves performance over monolingual models and enables cross-lingual speech-to-speech retrieval.
In this paper, we explore the learning of neural network embeddings for natural images and speech waveforms describing the content of those images. These embeddings are learned directly from the waveforms without the use of linguistic transcriptions or conventional speech recognition technology. While prior work has investigated this setting in the monolingual case using English speech data, this work represents the first effort to apply these techniques to languages beyond English. Using spoken captions collected in English and Hindi, we show that the same model architecture can be successfully applied to both languages. Further, we demonstrate that training a multilingual model simultaneously on both languages offers improved performance over the monolingual models. Finally, we show that these models are capable of performing semantic cross-lingual speech-to-speech retrieval.