BLAT: Bootstrapping Language-Audio Pre-training based on AudioSet Tag-guided Synthetic Data

Compared with ample visual-text pre-training research, few works explore audio-text pre-training, mostly due to the lack of sufficient parallel audio-text data. Most existing methods incorporate the visual modality as a pivot for audio-text pre-training, which inevitably induces data noise. In this paper, we propose BLAT: Bootstrapping Language-Audio pre-training based on Tag-guided synthetic data. We utilize audio captioning to generate text directly from audio, without the aid of the visual modality so that potential noise from modality mismatch is eliminated. Furthermore, we propose caption generation under the guidance of AudioSet tags, leading to more accurate captions. With the above two improvements, we curate high-quality, large-scale parallel audio-text data, based on which we perform audio-text pre-training. Evaluation on a series of downstream tasks indicates that BLAT achieves SOTA zero-shot classification performance on most datasets and significant performance improvement when fine-tuned on downstream tasks, suggesting the effectiveness of our synthetic data.

Can Audio Captions Be Evaluated with Image Caption Metrics?

Automated audio captioning aims at generating textual descriptions for an audio clip. To evaluate the quality of generated audio captions, previous works directly adopt image captioning metrics like SPICE and CIDEr, without justifying their suitability in this new domain, which may mislead the development of advanced models. This problem is still unstudied due to the lack of human judgment datasets on caption quality. Therefore, we firstly construct two evaluation benchmarks, AudioCaps-Eval and Clotho-Eval. They are established with pairwise comparison instead of absolute rating to achieve better inter-annotator agreement. Current metrics are found in poor correlation with human annotations on these datasets. To overcome their limitations, we propose a metric named FENSE, where we combine the strength of Sentence-BERT in capturing similarity, and a novel Error Detector to penalize erroneous sentences for robustness. On the newly established benchmarks, FENSE outperforms current metrics by 14-25% accuracy. Code, data and web demo available at: https://github.com/blmoistawinde/fense

Enriching Ontology with Temporal Commonsense for Low-Resource Audio Tagging

Audio tagging aims at predicting sound events occurred in a recording. Traditional models require enormous laborious annotations, otherwise performance degeneration will be the norm. Therefore, we investigate robust audio tagging models in low-resource scenarios with the enhancement of knowledge graphs. Besides existing ontological knowledge, we further propose a semi-automatic approach that can construct temporal knowledge graphs on diverse domain-specific label sets. Moreover, we leverage a variant of relation-aware graph neural network, D-GCN, to combine the strength of the two knowledge types. Experiments on AudioSet and SONYC urban sound tagging datasets suggest the effectiveness of the introduced temporal knowledge, and the advantage of the combined KGs with D-GCN over single knowledge source.