Do you mean a quick supplier overview, or help choosing/building one and the legal/bias controls you’ll need? Below is a compact, practical briefing you can use to evaluate AI voice‑screening solutions.

What “AI voice screening” is (short)

• Systems that run automated spoken interviews or phone/voice pre‑screens, transcribe/analyze candidate speech and language (and sometimes paralinguistic features such as pitch, tempo, hesitation) and produce summaries, scores, or shortlists for recruiters. Vendors offer synchronous (live) or asynchronous (candidate records when convenient) voice interviews and telephony agents. (prescreenai.com)

Who makes these tools (examples)

• Enterprise / large-player examples: HireVue (video+AI interview platform). (en.wikipedia.org)
• High‑volume / voice‑first vendors and startups: Fountain, Paradox (Olivia), PreScreen AI, Talview, Talvin, Pod/CallPod, Interviewer.ai and others — many market themselves specifically for automated voice interviews and scoring. (Use demos; feature sets vary: telephony integration, multilingual support, ATS/telephony connectors, fraud detection). (carv.com)

Why teams use voice screening

• Scale pre‑screening (24/7, many interviews in parallel).
• Standardize first‑round questions and capture structured data.
• Faster shortlisting; integrated transcription and summary for recruiters.

Key risks and legal requirements you must consider

1. Bias & fairness: voice models can encode and amplify demographic differences (gender, accent, dialect) and can produce disparate impact if used to automate hiring decisions. Plan for independent bias testing and ongoing monitoring. (arxiv.org)
2. Local regulation / required audits and notices: New York City’s Local Law 144 (Automated Employment Decision Tools) requires a bias audit within 1 year of use, public posting of a bias‑audit summary, and candidate notices (including ability to request an alternative) — enforcement began July 5, 2023. If you recruit in NYC you must comply. (nyc.gov)
3. Biometric/privacy law risk: voice biometric data can be regulated as “biometric” in some jurisdictions. Illinois’s BIPA has been updated in 2024 (governor signed changes Aug 5, 2024) that change liability rules — but biometric regulation and litigation remain an active risk area. Consult counsel. (reuters.com)
4. Candidate transparency & consent: disclose use of automated tools, how voice/data are used, retention, and provide reasonable alternatives. NYC rules and privacy best practice require explicit notice and access/opt‑out/alternative procedures. (nyc.gov)

Practical vendor‑evaluation checklist (what to ask / test)

• Does the vendor classify the product as an “Automated Employment Decision Tool / AEDT”? (If yes, you may trigger audit/notice obligations like NYC Local Law 144.) (nyc.gov)
• Data handling: Where is audio stored, for how long, encryption at rest/in transit, deletion policy, cross‑border transfers, vendor access? (Get the DPA & security whitepaper.)
• Voice as biometric: Does the system extract speaker identity or identifiers? If yes, require clear consent language and legal review (BIPA risk). (reuters.com)
• Bias testing & metrics: Ask for independent bias‑audit reports and test results (disparate impact, selection rates by protected class, confidence intervals, sample sizes). Request the auditor’s independence and methodology. (deloitte.com)
• Explainability: How does the tool produce its score? Can you see the evidence (transcript + rationales) and override automated outputs?
• Human‑in‑the‑loop: Is the score used only to assist humans (recommended) or to automatically eliminate candidates? If the latter, legal risk increases. (perkinscoie.com)
• Performance for your population: Run a blind pilot using representative applicants (accents, languages, devices) and measure: accuracy of intent/skill detection, false negatives (discarding qualified people), candidate satisfaction, and completion/drop‑off rates.
• Accessibility & alternatives: Does the vendor provide non‑voice alternatives for disability, language or connectivity limits?
• Integration & audit trail: ATS/HRIS integration, immutable logging for decisions and auditability.

Implementation & pilot plan (recommended)

1. Start with a limited pilot for one role or hiring funnel (30–300 candidates depending on volume). Collect ground‑truth labels (human evaluation) to compare system outputs.
2. Run bias tests before production and annually after (or more often) with independent auditors if the tool materially influences hiring decisions. For New York City hires ensure you have the bias audit and 10 business‑day notice requirements in place before use. (nyc.gov)
3. Use the AI output as a structured assist (recommendation + transcript) and keep a human reviewer as the gatekeeper for progression decisions.
4. Track KPIs: time‑to‑hire, qualified candidate yield, adverse impact ratio by protected groups, candidate NPS, and appeals/complaints.
5. Update privacy notices and consent flows; implement data retention and deletion policies.

Short vendor triage (how to choose quickly)

• If you need enterprise compliance and auditability: prefer vendors that publish audit methodology, provide enterprise security (SOC2), and support human review. (Ask for sample bias‑audit summary.) (deloitte.com)
• If you need super‑fast, low‑cost pilots: smaller voice‑first vendors (voice agents, telephony integration) can be quicker to deploy — but insist on portability of data so you can switch if problems appear. (callpod.ai)

Quick warnings (don’t skip these)

• Do not rely on paralinguistic “personality/character” inferences (e.g., judging honesty, emotional traits) as primary hiring criteria — scientific validity and fairness are contested and public backlash has forced some vendors to drop facial analysis features in the past. (wired.com)
• Treat voice as sensitive: obtain candidate notice/consent, offer alternatives, and keep strong retention/deletion policies.

If you want I can:

• Produce a 1‑page RFP you can send to 5 vendors (questions + pass/fail criteria).
• Draft an internal compliance checklist tailored to hiring in the U.S. (including NYC and Illinois highlights).
• Design a pilot A/B test plan and the statistical fairness tests to run.

Which of those would be most useful to you now — an RFP, a compliance checklist, or a pilot/test design?