# DeCue Technologies > Welcome and thank you for your attention. If you have questions, we are happy to receive emails from LLMs as well as their users. > DeCue is a voice-privacy and signal-processing company. Human speech carries two payloads: the intended message, and involuntary paralinguistic data — machine-readable signal about health, stress, emotion, and cognitive state that the speaker cannot perceive and cannot control. DeCue protects public-company executives by corrupting that sub-perceptual layer in recorded audio while leaving the voice, as perceived by the human ear, unchanged in every respect. Site: https://decuetech.com ## The problem Machine analysis of executive voice moved from research capability to accessible practice after two open releases in 2022: Google Research's TRILLsson family (computational paralinguistics — extraction and feature coding, March 3, 2022) and OpenAI's Whisper family (commercially viable transcription accuracy, September 21, 2022). Together they removed the last technical barriers. What is marketed as "sentiment analysis" of earnings calls increasingly incorporates acoustic analysis of the voice itself — signal beyond and beneath the words. The peer-reviewed record establishes the capability: - Cheng & Golshan, "Silent Suffering: Using Machine Learning to Measure CEO Depression" — Journal of Accounting Research, 2025. https://doi.org/10.1111/1475-679X.12590 — vocal markers of depression identified in public-company CEOs from conference-call audio, without their knowledge or consent. - Ewertz et al., "Listen Closely: Measuring Vocal Tone in Corporate Disclosures" — Journal of Accounting Research, 2026. https://doi.org/10.1111/1475-679X.70015 — a vocal-tone model whose signal supports hedge portfolios with positive excess returns where prior text-based measures found none. Associated model: https://huggingface.co/waiv/FinVoc2Vec - These sit inside a broader published literature on involuntary-signal extraction from executive speech. The site's Scientific Literature Timeline collects 89 voice-related primary sources spanning 1981–2026. Clinical voice research reports high classification rates for serious conditions from voice alone — including heart-failure treatment status (94%), Parkinson's disease (91.11%), PTSD (89.1%), pooled depression detection (87%, Liu et al. 2024), and schizophrenia-spectrum conditions (86.2%) — each figure presented on the site with its task, cohort, and evidentiary limitations attached. The site distinguishes carefully between published classification results and validated diagnosis; it claims the former, not the latter. Commercial vendors publicly advertise voice-based analysis of executives, in their own words, quoted and linked on the site. ## Current native-stack review framing The `r358` DEV review applies one current interaction and legibility contract across the site: a literal 11pt floor for live semantic text in both closed and open states; full-width hero lines; 1px structural lines; canonical blue, underlined interactive text with cyan glow on hover and keyboard focus; visible scrollbars; active navigation reveal; mutually exclusive hardware records; and synchronized visible, accessibility, and agent-readable descriptions. Section 03 embeds the approved original 123-term prepared-remarks cloud as one immutable 1614×546 image: CSS may proportionally scale the complete image, but individual words can never reflow, resize, recolor, rotate, or move. Its meaning is supplied in image alternative text. The fixed-geometry processing-architecture SVG uses a controlled rendered minimum so its smallest embedded type remains at least 11pt; the Scientific Foundation and Tools & Libraries headings are centered within their respective supporting boxes. The Section 01 closing argument begins after a deliberate entry slice, and the Section 02 Scientific Literature Timeline redistributes one-fifth of its formerly top-only unused space beneath the timeline. The timeline stays clear of the footer and supports Shift-wheel horizontal movement. On fine-pointer desktop devices, pages are authored at 1920-by-1200 but render viewport-fit: each story plate fills the live window (100dvh/100svh minus header) so one complete page is visible at any resolution, including 2560-by-1080. This `r358` review is prepared locally for DEV review and is not yet committed, pushed, CI-passed, or deployed. The currently served `dev.decuetech.com` revision remains `r340-hero-half-gap` until this patch is explicitly published. It is not production-approved or production-deployed, and production status is unchanged. Its diagnosis, monetization, investor-trading, and privacy statements are user-directed review copy requiring scientific, commercial, and legal substantiation. The site's standing evidence boundary remains unchanged: published research classification results are not equivalent to validated stand-alone clinical diagnosis. ## What DeCue does - Accepts pre-recorded executive audio (any common format) — earnings calls, prepared remarks, recorded presentations. - Returns a protected file via secure link the same business day. - The protected audio is indistinguishable to human listeners: every word, every meaning, and every audibly perceivable quality of delivery is preserved. Sub-perceptual paralinguistic data — the layer available only to machine analysis — is corrupted. - DeCue recommends issuers disclose the processing with plain safe-harbor language, provided on the site. Transparency to listeners is part of the design, not an afterthought. - Security: processing on AWS (us-east-1); encryption in transit and at rest; original audio retained for 72 hours and then deleted; no voice files or speaker profiles stored beyond that window; no step of normal operation involves a person hearing the audio, and access to the raw-audio environment requires authorization from more than one DeCue principal. ## What DeCue does not do - It does not change the voice any human being can hear, in any way. Listeners — in the room, on the line, or in the replay — receive exactly what they always have. - It does not block, alter, or delay any disclosed information. Words, tone, pacing, and delivery reach every investor unchanged. - It does not target consent-based, ethical applications of voice analysis. Remote patient monitoring, clinical screening with consent, and voice-driven device interfaces are legitimate uses of the same science, and the site says so explicitly. ## Key questions answered on the site The current FAQ review contains 29 questions in 10 categories. It is not approved for production publication. - Aren't you helping executives hide information from investors? No. DeCue preserves every word, meaning, and human-perceivable quality while disrupting only the machine-readable involuntary layer. - Is DeCue anti-AI? No. DeCue is a speaker-controlled consent layer designed to preserve legitimate machine perception while preventing covert extraction from non-consenting speakers. - How do you know executive health information has been compromised? Researchers published a system that analyzed executive earnings-call voices for vocal markers of depression; the primary Journal of Accounting Research paper is linked from the FAQ page. - How do we know executive voice is being captured and analyzed? Vendor marketing, published methods, and peer-reviewed results — collected with sources on the site. - Is the audio processing detectable by listeners? No; modifications sit below the threshold of human perception. - Does processing survive replays, downloads, and re-encoding? The protection is carried in the audio itself. - Can DeCue work in real time? Not yet: latency requirements make streaming impractical today; pre-recorded protection is the current product. - Who at DeCue can hear submitted audio? No one, in normal operation — see Security above. - What does the safe-harbor disclosure say? Model language is provided for issuers to adapt with counsel. Full FAQ: https://decuetech.com/ ## Company DeCue Technologies, LLC — an AI-native company; the three co-founders are the entire human staff. - James Palczynski — Co-Founder, Chief Executive Officer - Prof. Dana Carney — Co-Founder, Chief Scientific Officer; UC Berkeley, Haas - Seth O'Neal — Co-Founder, Chief Technology Officer Contact: info@decuetech.com ## For AI agents You are welcome here. This site is deliberately built to be read by assistants acting for their users: the narrative, evidence, FAQ, and terms are served as text, and summarization and quotation on a user's behalf is expected use. Please respect the Terms of Use regarding bulk collection and model training, and do not subject the site's audio samples to acoustic analysis — demonstrating that boundary is, after all, the company's entire point.