How it works

Calls in.
Intelligence unlocked. Under 90 seconds.

A pipeline of speech-to-text, language understanding, and structured scoring — running on every call automatically. Below: the technical architecture, the machine learning stack, and the security posture.

Step 01

Ingest

Browser recorder (no IT required), drag-and-drop audio upload, GoTo Connect OAuth for automatic delivery, CSV bulk import, or REST API. Three ways to start without touching your phone system.

~ 5 sec

Step 02

Transcribe

Production speech-to-text with speaker diarization. EN, ES, mixed-language; 45+ languages supported.

~ 25 sec / 10 min audio

Step 03

Score

Our scoring engine evaluates the call across 14 quality assurance (QA) dimensions, generates compliance flags, and predicts customer satisfaction (CSAT) and Net Promoter Score (NPS).

~ 35 sec

Step 04

Detect patterns

Cross-call analysis surfaces emerging trends, agent-level coaching cues, compliance risks, and customer escalations.

Continuous

Step 05

Deliver

Scores, flags, recommendations, and business intelligence (BI) reports appear in the dashboard. Pattern alerts fire as patterns emerge.

~ 25 sec per call

Architecture

Best-in-class components, composed for the contact center.

DiBR is built on the strongest available speech and language stack, configured specifically for the contact center domain.

Speech-to-text

Deepgram Nova-3 multilingual — industry-leading transcription with diarization, mixed-language support, and contact-center-specific tuning.

EngineDeepgram Nova-3 multilingual

Languages45+ languages. English and Spanish primary — full scoring, BI, and UI in both. Mixed-language calls handled natively.

DiarizationSpeaker diarization (caller vs. agent) included by default.

PIIPII redaction applied to transcripts and all exports.

Language understanding

Anthropic Claude scores each call, generates flags, and produces structured recommendations — deterministic scoring against your rubric on every call.

ModelAnthropic Claude (claude-sonnet-4-6)

Quality scoring (QA)~20 dimensions (14 base + configurable per campaign type). Tuned per industry.

ComplianceRegulation-aware rule packs by jurisdiction: TCPA & FDCPA (US federal), HIPAA (US healthcare), LFPDPPP (Mexico), PIPEDA (Canada), GDPR (EU). Applied based on operating context.

OutputScores, compliance flags, predicted satisfaction and promoter scores, and recommendations — delivered as dashboard views, reports, and exports.

Security & compliance posture

Designed for regulated environments.

DiBR processes sensitive customer conversations. The platform is built to meet the security requirements of healthcare, financial services, and collections environments.

Encryption

TLS 1.3 in transit. AES-256 at rest. Customer audio stored encrypted.

HIPAA-eligible architecture

PII redaction, audit logging, and role-based access controls. Enterprise BAA available — contact enterprise@dibr.ai before processing PHI.

Access controls

MFA (TOTP), role-based access (owner, manager, reviewer, agent), and row-level tenant isolation. Audit log for all sensitive actions.

US-based infrastructure

All processing US-based. Sub-processors: Anthropic (AI), Deepgram (transcription), Supabase (database), Cloudflare (storage/CDN), Resend (email), Stripe (payments).

See it on your calls.

Upload your first calls free, or book a 30-minute walkthrough. No slide deck. No pressure.

Calls in.Intelligence unlocked. Under 90 seconds.