Glossary

Voice AI in plain language.

A voice AI agent is software that answers calls with a synthetic voice, understands speech and routes or calls back a human — without the push-button menus of an IVR. This glossary defines the key terms of voice AI, GDPR, the Italian AI Act and B2B telephony for anyone evaluating adoption.

Voice AI agent

Software that answers phone calls with a synthetic voice, understands speech, captures structured information, and routes the conversation to a human or schedules a callback. Unlike an IVR, it doesn't require button menus.

IVR (Interactive Voice Response)

Touch-tone phone system ("press 1 for…"). A 1990s technology, now replaceable by a voice AI agent that performs natural-language intent detection.

Voice AI

Umbrella term for AI models that recognise and generate human speech in real time. Typical stack: STT (speech-to-text) → LLM (intent) → TTS (text-to-speech).

TTS — Text-to-Speech

Voice synthesis: converts written text into spoken audio. In 2026 native-quality voices are available through the major cloud platforms; SmartVolve uses the Google Cloud (Vertex AI) voice stack in the EU region.

STT — Speech-to-Text

Speech recognition: converts spoken audio into text. Modern models (Whisper, Deepgram) achieve >95% accuracy on conversational English and Italian.

RAG — Retrieval-Augmented Generation

Technique that lets an LLM answer using company-specific content (FAQs, price list, catalog) retrieved on the fly. Reduces hallucinations because the AI cites from the knowledge base instead of inventing answers.

Warm handoff

Transferring a call from the AI agent to a human with context already captured (name, reason, urgency). Reduces friction for the end caller compared to a cold blind transfer.

SIP trunk

VoIP line that connects a business phone system to the public telephone network. It's the typical entry point for plugging a voice AI agent in without changing the existing phone number.

DPIA — Data Protection Impact Assessment

Privacy impact assessment required under GDPR Art. 35 for large-scale processing or novel technologies — voice AI agents qualify. Identifies risks, mitigation measures, and residual risk.

DPA — Data Processing Agreement

Contract between data controller and data processor required by GDPR Art. 28. Defines purpose, duration, data categories, security measures, and authorised sub-processors.

Sub-processor

Technology vendor that processes personal data on behalf of the main processor. Examples: cloud (Google Cloud), email delivery (Resend), CRM (Airtable). Must be disclosed in the DPA and notified to clients.

AI Act art. 50

Article of the EU Artificial Intelligence Act (2024) that requires disclosing the synthetic nature of an AI voice at the start of every conversation with a human interlocutor.

EU data residency

Physical server location inside European territory (e.g. Google Cloud, EU region). Simplifies GDPR compliance because SCCs and a Transfer Impact Assessment are not required for intra-EU processing.

Pseudonymisation

GDPR technique (Art. 4 §5) that replaces direct identifiers (name, email) with codes that cannot be linked back to the individual without additional information kept separately. SmartVolve pseudonymises transcripts after 30 days.

Audio AI watermarking

Inaudible marker embedded in synthetic audio to identify it as AI-generated. Still evolving in 2026; the major providers are rolling it out as an anti-deepfake measure.

Double opt-in

Newsletter sign-up flow that requires email confirmation before any marketing communication is sent. Standard required by EU data protection authorities and recommended under US anti-spam rules.