Why Open Voice Shield
Why Open Voice Shield: the one-page case
Open Voice Shield is one SIP hop on the trunk you already run. Every call comes back with a fraud verdict and its evidence, a route alert if the vendor cheated, and the audio quality of each leg. A flagged number's next call is blocked or diverted automatically. Nothing is installed on your switch, and a three-minute call costs $0.08.
The problem
The exposure is per call
Two kinds of loss arrive one call at a time, and nothing in a typical voice stack hears either of them happen.
Calls your customers make. A scam or an unlawful telemarketing call carries statutory exposure per communication: under the Telephone Consumer Protection Act, $500 to $1,500 per illegal call, uncapped. A non-compliant campaign that runs for a day has already generated a year's worth of damages before anyone reviews it.
Calls you pay for. A vendor route that answers on ringback, or swaps the far end to a recording a minute in, bills minutes nobody spoke. Manual test calls do not see it because the fraud is timed to outlast them.
CDRs show where calls went and for how long. Caller ID attestation shows who was allowed to present a number. Neither shows what was said, which is where both losses are decided.
The product
One hop, three things on every call
- A verdict with evidence. Probability, category, the entity the caller impersonated, numbered red flags tied to what was said, a TCPA compliance note, and a block, review or allow recommendation, with the stereo recording and speaker-labelled transcript behind it.
- A route alert. Early answer on ringback, measured after the 200 OK, and delayed playback, read from the callee channel, reported separately from the caller verdict so it reaches the carrier relationship, not the fraud desk.
- Call quality per leg. Estimated MOS, jitter and packet loss for the caller leg and the callee leg, from the same capture, so a degrading vendor shows on the far leg while your side stays clean.
- Enforcement on the next call. Automation rules block or divert a flagged number for the window you choose. Alerts go by e-mail or to your Slack channel, and a signed webhook carries every verdict to your own systems.
- Evidence you can hand over. A branded PDF report, an offline export bundle with recording, capture and SIP log, and a password-protected share link.
- Operable by software. Everything above is typed data on the API, and an AI agent can run the account through the MCP server, from the OAuth sign-in to a test call through your route, which you confirm, to the verdict.
Positioning
How it differs from what you already run
Compared by approach rather than by vendor. Most of these are complementary, and none of them hears the call. Comparing against a named product? See the VeriSwitch SmartShield alternative and FraudCentral VoiceShield alternative pages.
| Approach | What it sees | What it cannot do | Open Voice Shield |
|---|---|---|---|
| CLI reputation, STIR/SHAKEN attestation | The number and its attestation, at signalling time. | Hear the conversation. A well-attested number can still run a scam. | Reads what was said and acts on the number's next call. Keep attestation and add the verdict. |
| Call analytics and QA platforms | Recordings your own systems made, scored for quality or sales outcomes. | Record on any trunk, judge fraud and compliance, check the route, enforce. | Makes the recording itself as a SIP hop, scores fraud and TCPA, grades the route and the audio, blocks or diverts. |
| CDR-based fraud management | Destinations, volumes, durations, patterns, after the fact. | Catch one call on its content, or explain why a call was bad. | Catches a single scam call on the words, with the red flags as the reason. |
| Manual route testing | A few short test calls a day, by hand. | See delayed playback timed to outlast the test, or test every route every hour. | Reads every sampled live call, and an agent can test every route on a schedule with typed results. |
| An SBC or firewall | Signalling and media at the edge, by rule. | Understand the call. Rules need someone to know the bad number first. | Produces the bad number, with evidence, for the rule to act on. |
It does not replace a B2BUA or an SBC, and it is not a campaign dialler. In standard mode verdicts are produced after hangup and enforcement is the next call from that number. Realtime mode, set up on request, scores the call while it is still up.
Proof, and what we do not claim
Judge it on your own traffic
The technical white paper describes the threat model, the call path, every pipeline stage and the measurement method. Latency and transcription error were measured on a scripted corpus, and the method, corpus and raw results are shared on request. The 63-second film and the six-minute tour show the real product on a real scam call. The documentation is written for the engineer doing the work, with every value taken from the running system.
We publish no accuracy, precision or recall figure, because none has been computed against a labelled corpus of real calls. The verdict is an analyst's read with its evidence attached, not a legal finding, and the TCPA note is a reading of the conversation, not legal advice. Estimated MOS is an estimate from loss and jitter, never a measured opinion score.
To evaluate it, route one call in ten through the hop for a week, read the verdicts against what your team already knew, and count the route alerts.
What is always true on every call
- Nothing is injected into the live audio, and analysis runs after the BYE.
- The ingress fails closed: if we are unreachable, your switch takes its normal alternate route.
- Every call keeps its recording, capture and SIP ladder as evidence.
- Recordings live in storage we control, with retention that deletes the object, and every endpoint is scoped to one tenant and audited.
- Analysis can run entirely on our own servers, with no third-party AI involved.
Who uses it
Three teams and the job each does
Carrier and wholesale NOC
Sample live traffic through the hop and catch early-answer and delayed-playback FAS on vendor routes with the evidence to take to the vendor. Let an agent test every route every hour. FAS detection.
Contact centre and compliance
A TCPA note on every analyzed call minutes after hangup, alerts when a campaign crosses the line, and the transcript counsel asks for. TCPA penalty prevention.
CPaaS, CCaaS and UCaaS platforms
Put a fraud verdict and a quality score on every call your customers make, by API and webhook, with automation on the next call. Built for agents and the API.
Commercials
What it costs
The price is analyzed minutes at the plan rate plus one analysis fee per call, prepaid. On the default plan a three-minute checked call is $0.08, with recording, transcription, analysis and evidence included. There are no seats and no monthly minimum on pay-as-you-go, and $5 of credit at signup means the first calls need no card. Bundled plans trade a monthly fee for a lower rate. The pricing page reads its figures from the live plans API.
Next step
Put one call through it
Whitelist an IP, point a trunk, place one test call: the first verdict lands the same afternoon. Or hand the whole setup to an agent.
Questions
Frequently asked questions
How is Open Voice Shield different from STIR/SHAKEN or CLI reputation?
Those work at signalling time on the number and its attestation. They never hear the call. Open Voice Shield reads the conversation after hangup and returns a verdict with the words that earned it, then blocks or diverts that number on its next call. The two are complementary: attestation says who is allowed to present a number, and the verdict says what they did with it.
How is it different from call analytics or QA platforms?
Analytics platforms transcribe recordings your own systems made and score them for quality or sales outcomes. Open Voice Shield is a SIP hop: it makes the recording itself on any trunk, scores the call for fraud and compliance, checks the route for false answer supervision, grades the audio per leg, and enforces on the next call. It needs nothing installed on your switch.
How is it different from carrier fraud management systems?
Those read call detail records: destinations, volumes, durations and patterns, after the fact. They never hear what was said. Open Voice Shield reads the conversation of each call it carries, so a single scam call is caught on its content instead of after a pattern forms.
What does it cost?
Per analyzed minute at the plan rate plus one analysis fee per call, prepaid. On the default plan a three-minute checked call is $0.08 including recording, transcription, analysis and evidence. Signup carries $5 of credit. There are no seats and no monthly minimum on pay-as-you-go.