Each language has a default voice model. Most languages default to ElevenLabs V3 conversational, and Japanese defaults to Kotoba. To have your Agentforce Voice agent speak with a different voice model and voice persona, use the modality voice block of Agent Script.
Where: This change applies to Lightning Experience in Enterprise, Unlimited, and Developer editions with Foundations or Agentforce 1 editions, and Salesforce Voice add-ons.
When: Voice model selection is available starting the week of August 24, 2026.
Why: Before this change, an Agentforce Voice agent could speak only with its
language's default voice model. Select a lower-latency Flash model for faster spoken
responses, gain finer control over the speed, stability,
and similarity parameters, improve handling of European and Australian
English accents, or choose a Japanese voice with Kotoba. Agentforce Voice supports
ElevenLabs V3 conversational (multilingual), ElevenLabs Flash V2.5 (multilingual),
ElevenLabs Flash V2 (English only), and Kotoba (Japanese).
How: In the outbound section of the modality
voice block, set model.id to the model you want and set
persona_id to a voice from that model's voice catalog. Omit the
model to keep the language's default. Set a non-default voice model in Script view or
Agentforce DX. The Canvas view doesn't expose a voice-model selector. Existing scripts
that use the legacy voice format continue to work without changes.