IPA-CLIP: Integrating Phonetic Priors into Vision and Language Pretraining
This addresses a specific limitation in vision-language models for multimedia systems, but it is incremental as it builds on existing CLIP architecture.
The paper tackles the problem that vision-language pretrained models like CLIP ignore pronunciation, which humans use to understand language, by integrating phonetic priors into CLIP to consider pronunciation similarity. The result shows that the proposed IPA-CLIP model improves phoneme representation accuracy and enhances performance in multimodal retrieval tasks, handling nonsense words more phonetically.
Recently, large-scale Vision and Language (V\&L) pretraining has become the standard backbone of many multimedia systems. While it has shown remarkable performance even in unseen situations, it often performs in ways not intuitive to humans. Particularly, they usually do not consider the pronunciation of the input, which humans would utilize to understand language, especially when it comes to unknown words. Thus, this paper inserts phonetic prior into Contrastive Language-Image Pretraining (CLIP), one of the V\&L pretrained models, to make it consider the pronunciation similarity among its pronunciation inputs. To achieve this, we first propose a phoneme embedding that utilizes the phoneme relationships provided by the International Phonetic Alphabet (IPA) chart as a phonetic prior. Next, by distilling the frozen CLIP text encoder, we train a pronunciation encoder employing the IPA-based embedding. The proposed model named IPA-CLIP comprises this pronunciation encoder and the original CLIP encoders (image and text). Quantitative evaluation reveals that the phoneme distribution on the embedding space represents phonetic relationships more accurately when using the proposed phoneme embedding. Furthermore, in some multimodal retrieval tasks, we confirm that the proposed pronunciation encoder enhances the performance of the text encoder and that the pronunciation encoder handles nonsense words in a more phonetic manner than the text encoder. Finally, qualitative evaluation verifies the correlation between the pronunciation encoder and human perception regarding pronunciation similarity.