TTS (Text-to-Speech) is the agent's voice. Shopbell ships with Kokoro TTS — open-source, free, human-sounding. Four voices are available. Each has a name, a personality, and a typical use case. This guide covers each voice and how to configure.
Bella (af_bella) — Lead voice, recommended default
Female, warm, clear, professional. Use for: most inbound calls (dental, salon, retail, restaurant). The 'florist who answers the phone' vibe. Friendly but not casual. Pairs well with GPT-4o-mini for cost-effective friendly receptionists.
Adam (am_adam) — Alternative male lead
Male, calm, measured, authoritative. Use for: legal intake, financial services, B2B sales, anywhere a female voice might sound too young for the audience. Pairs well with Claude 3.5 Sonnet for compliance-sensitive agents. Heard in our regenerated industry sample audios.
Sky (af_sky) — Customer support, technical
Female, slightly tech-adjacent, concise. Use for: SaaS support lines, technical product inquiries, anywhere the caller expects quick answers and less warmth. Pairs well with GPT-4o for high-bandwidth reasoning.
Nicole — RETIRED
Female, but the model sounded sleepy on long calls. We deprecated Nicole and recommend Bella instead. Existing workflows using Nicole will fail at TTS time — switch to Bella or Sky. If you have an old workflow using Nicole, edit the TTS configuration in the workflow's Settings, save, test.
Setting the voice in your workflow
Open the workflow. Click any Agent node. In the right-side panel, find the TTS Voice dropdown. Pick Bella, Adam, Sky (Nicole disabled). Save. Test the workflow with a Web Call — you'll hear the new voice immediately on the greeting. Every Agent node can have its own voice if you want different voices at different stages of the conversation (rare — usually one voice per workflow).
Frequently Asked Questions
Can I use ElevenLabs or OpenAI TTS instead of Kokoro?
Yes via BYOK. Click Model Configurations, New Configuration, TTS provider, paste your ElevenLabs or OpenAI TTS API key. Then assign that configuration to your workflow's TTS dropdown. Kokoro remains the free default.
Are Spanish / other languages supported?
Kokoro currently supports English. For Spanish or other languages, BYOK with OpenAI TTS (multi-language) or ElevenLabs. We plan to add Kokoro multilingual voices as they become available upstream.
Can the agent switch voices mid-call?
Yes but rare. Each Agent node can use a different TTS voice. Set up the workflow with Agent A using Bella for greeting, Agent B using Adam for technical support, etc. Most workflows use one voice throughout.