Current browse context:
cs.CL
Change to browse by:
References & Citations
Computer Science > Computation and Language
Title: Homophone Reveals the Truth: A Reality Check for Speech2Vec
(Submitted on 22 Sep 2022 (v1), last revised 23 Sep 2022 (this version, v2))
Abstract: Generating spoken word embeddings that possess semantic information is a fascinating topic. Compared with text-based embeddings, they cover both phonetic and semantic characteristics, which can provide richer information and are potentially helpful for improving ASR and speech translation systems. In this paper, we review and examine the authenticity of a seminal work in this field: Speech2Vec. First, a homophone-based inspection method is proposed to check the speech embeddings released by the author of Speech2Vec. There is no indication that these embeddings are generated by the Speech2Vec model. Moreover, through further analysis of the vocabulary composition, we suspect that a text-based model fabricates these embeddings. Finally, we reproduce the Speech2Vec model, referring to the official code and optimal settings in the original paper. Experiments showed that this model failed to learn effective semantic embeddings. In word similarity benchmarks, it gets a correlation score of 0.08 in MEN and 0.15 in WS-353-SIM tests, which is over 0.5 lower than those described in the original paper. Our data and code are available.
Submission history
From: Guangyu Chen [view email][v1] Thu, 22 Sep 2022 05:32:09 GMT (908kb,D)
[v2] Fri, 23 Sep 2022 11:10:15 GMT (908kb,D)
Link back to: arXiv, form interface, contact.