Salesforce unveils Launa-Omni voice model to boost enterprise AI responsiveness

Salesforce AI Research has developed Launa-Omni, a new voice model designed to offer real-time support for two-way spoken conversations in enterprise applications like Salesforce and Slack.

The architecture integrates audio processing directly into a core language model while utilizing an in-house text-to-speech component to preserve speech characteristics like pace and hesitation.

Launa-Omni achieved competitive benchmark scores, including 90.60 per cent on VoiceBench and a macro score of 72.9 per cent on the τ³-Voice customer-service evaluation.