English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Automatic speech recognition and voice identification systems are beingdeployed in a wide array of applications, from providing control mechanisms todevices lacking traditional interfaces, to the automatic transcription ofconversations and authentication of users. Many of these applications havesignificant security and privacy considerations. We develop attacks that forcemistranscription and misidentification in state of the art systems, withminimal impact on human comprehension. Processing pipelines for modern systemsare comprised of signal preprocessing and feature extraction steps, whoseoutput is fed to a machine-learned model. Prior work has focused on the models,using white-box knowledge to tailor model-specific attacks. We focus on thepipeline stages before the models, which (unlike the models) are quite similaracross systems. As such, our attacks are black-box and transferable, anddemonstrably achieve mistranscription and misidentification rates as high as100% by modifying only a few frames of audio. We perform a study via AmazonMechanical Turk demonstrating that there is no statistically significantdifference between human perception of regular and perturbed audio. Ourfindings suggest that models may learn aspects of speech that are generally notperceived by human subjects, but that are crucial for model accuracy. We alsofind that certain English language phonemes (in particular, vowels) aresignificantly more susceptible to our attack. We show that the attacks areeffective when mounted over cellular networks, where signals are subject todegradation due to transcoding, jitter, and packet loss.

Hear "No Evil", See "Kenansville": Efficient and Transferable Black-Box Attacks on Speech Recognition and Voice Identification Systems