Salesforce Launches Launa-Omni for Enterprise Voice Tasks
The new model generates visible text before spoken audio to improve interpretability in business applications.
Updated on Oct. 21, 2026 in Artificial Intelligence
Live Poll
Do you trust AI-driven voice systems to handle your customer service inquiries accurately?
Salesforce AI Research has announced Launa-Omni, a new voice model designed to integrate audio processing directly into enterprise-focused language models. A simplified iteration of this technology is already being utilized in the Slackbot Voice Dictation feature.
Why it matters
The model aims to improve the responsiveness and accuracy of business voice applications by preserving nuanced speech patterns like pacing and hesitation. This development represents a shift toward more interpretable enterprise AI, as the system generates visible text responses prior to speech output.
Launa-Omni achieved 90.60% on the VoiceBench benchmark and a 72.9% macro score on the τ³-Voice benchmark. It performed at 59.69% on the Audio MultiChallenge, trailing the current top public score of 60.40%.
The players
Salesforce
An enterprise cloud computing company that builds customer relationship management software and proprietary AI models.
Slackbot
A productivity software assistant integrated into the Slack collaboration platform that now leverages a simplified version of Launa-Omni.
The details
Launa-Omni utilizes a dual-component architecture where audio processing is embedded within the core language model while speech generation is handled by a separate module. A real-time turn-taking system monitors speaker activity to determine when a user has finished speaking, triggering an immediate response. This design preserves expressive human nuances such as hesitation and emphasis that are often lost in standard text-to-speech pipelines.
Timeline
October 21, 2026: Salesforce AI Research published details regarding the Launa-Omni model.
The Tech Race
Launa-Omni enters a crowded field of voice-processing research currently measured by the VoiceBench benchmark. Salesforce aims to bridge the gap between academic performance and business-ready interpretability by prioritizing text-to-speech transparency.
Users of Slack can currently interact with a simplified version of this voice technology via the platform's native dictation features. Future business applications will likely rely on the full model's ability to display text before spoken responses to enhance accuracy and oversight.
The takeaway
The move toward dual-component voice models reflects a broader push for interpretability in enterprise AI tools. Observers should track subsequent integration announcements from Salesforce to determine if Launa-Omni reaches parity with current leading Audio MultiChallenge benchmarks.
Further reading
For broader trends in enterprise-grade models, see the latest developments in Artificial Intelligence.
Source note: This article includes information reported by The Indian Express.
Live Poll
Do you trust AI-driven voice systems to handle your customer service inquiries accurately?
