Using Sonic-3.6 for Customer Support Phone Agents and Voice Self-Service

On August 27, 2026, Cartesia introduced Sonic-3.6, the latest version of its text-to-speech model focused on enhancing the naturalness and quality of AI-generated speech for real-time applications. Sonic-3.6 is a generative audio model designed to produce highly realistic and contextually appropriate voice output across multiple languages and locales.

Compared to earlier versions, particularly Sonic-3.5, Sonic-3.6 makes a marked improvement in voice naturalness, with listeners preferring it in up to 93% of blind head-to-head tests across fifteen regions. This represents a significant leap beyond the voice quality available from prior TTS models and standard API-based text-to-speech systems commonly used in automated agents and contact center IVRs.

For support and contact-center teams, this release enables the creation of AI phone agents and self-service systems that are more credible and easier for callers to interact with, while lowering the effort needed to comply with regulatory requirements like clear disclosure and informed consent during voice interactions. The change makes it possible to automate more customer support calls without breaking the conversational trust or quality thresholds that previously forced escalation to live agents.

What Sonic-3.6 Can Do for Customer Support Teams

Sonic-3.6 enables contact centers and support teams to build AI phone agents and self-service systems that sound more natural and are easier for callers to understand and trust. The model’s improved speech clarity and prosody allow it to handle a wide range of call flows, including gathering caller intent, confirming information, providing account details, triaging issues, and reading regulatory disclosures.

Sonic-3.6’s broad locale coverage and preference in blind tests may reduce caller frustration associated with synthetic-sounding or robotic voices, which historically led to frequent escalations or early hang-ups.

• Automated call routing and intent capture
• Information lookup and repetitive task handling
• Reading required disclosures in a clear, listener-friendly tone
• Collecting caller responses for forms or surveys
• Basic troubleshooting and scripted decision trees

Call Handling and Escalation Scenarios

Sonic-3.6 can be used to automate standard call handling steps and escalate interactions when specific triggers are detected. Common escalation scenarios include requests for a human agent, statements that indicate caller distress, or situations where required consent cannot be confirmed via the automated flow.

Contact centers can configure call scripts and workflows to ensure the AI agent stays within its design limits and escalates appropriately according to business logic or compliance guidelines. Sonic-3.6’s natural-sounding output makes mixed-mode conversations—where a call may move from AI to a live agent—less jarring for the customer.

Consent and Disclosure Requirements in Voice Workflows

AI phone agents powered by Sonic-3.6 must be configured to meet legal requirements for consent, clear disclosure, and data collection in regulated industries. This includes informing callers they are interacting with an automated system, explaining how data will be used, and capturing explicit agreement where laws require it.

Because Sonic-3.6 outputs are rated as more natural and listener-friendly, support teams can tune the timing, tone, and repetition of mandatory disclosures to reduce misunderstanding and lower repeated consent requests—while still recording and logging consent for audits.

Integrating Sonic-3.6 in Contact Center Systems

To put Sonic-3.6 into production, teams can use Cartesia’s developer APIs and SDKs to connect the TTS model with existing IVR (Interactive Voice Response) platforms, customer relationship management (CRM) tools, or cloud contact-center suites. Cartesia supports voice agent development through its Line platform, and Sonic-3.6 can be paired with Cartesia’s Ink-2 speech-to-text model for two-way real-time conversations.

Support teams should validate compatibility with their call routing, logging, and compliance monitoring systems before rollout. Cartesia documentation recommends tuning configuration options for locale, voice profile, and disclosure scripting during implementation.

• API and SDK support for common telephony stacks
• Prebuilt voice profiles for major locales
• Logging and auditing integration for regulatory tracking
• Documentation and developer resources from Cartesia

Model Compliance and Data Handling Considerations

Support teams deploying Sonic-3.6 in regulated contexts need to confirm regulatory alignment, including caller consent, disclosure requirements, and data residency obligations. Cartesia’s prior compliance updates indicate that its text-to-speech platform is GDPR-compliant, but support teams in healthcare, finance, or other regulated sectors should review the latest documentation and verify any model-specific certifications before launch.

It is essential to log when a customer is given a disclosure, when they consent (or do not), and to record system prompts for audits or disputes.

Operational Insight: What Changes in Real Deployments

Based on the voice-automation systems we have implemented across support and content teams, the switch to a more natural-sounding model like Sonic-3.6 typically reduces the number of calls escalated to live agents solely due to voice quality complaints, and increases the acceptance rate of mandatory disclosures.

Most operational failures in AI voice agent rollouts happen when custom consent scripting or integration with call logging systems is skipped or done manually. Teams who run a pilot and analyze real consent capture rates before scaling site-wide avoid costly remediation after the fact.