- DATE:
- AUTHOR:
- The Krisp team
Krisp 2.79.2 Call Center AI Updates
This release brings more natural conversations with upgrades across Accent Conversion and Voice Translation, interaction visibility and stronger privacy in Speech Analytics through agent screen recording and audio PII redaction. Let's take a closer look.
Accent Conversion improvements
More natural Accent Conversion: The Filipino and LatAm English accent models got significant improvements in speech naturalness, pronunciation clarity, and voice preservation.
Agent Accent Conversion auto-switch based on CCaaS skills: For teams using Krisp CX Bridge, Accent Conversion can now turn on for English calls and off for others based on the call's CCaaS routing skill before the conversation even begins. Admins map skills to languages in the admin portal.
Voice Translation improvements
Manually interrupt translated speech: When a customer jumps in while the agent's translated speech is still playing, the agent can now instantly drop the translation playback to listen and respond right away, making the bilingual conversations feel as natural and responsive to customers as same-language ones.
Translated voices that match the speaker gender: Voice Translation now automatically matches the translated voice to the speaker's gender, making multilingual conversations sound more intuitive. For the agents, they are prompted to confirm their translation voice gender when they first activate the app.
Lower network usage in Voice Translation: Voice Translation now works better on slower or busy networks, helping translated calls stay smooth with less bandwidth.
Whisper tone for customer voice changes: When a new speaker is detected on the customer side, agents now hear a short whisper tone audible only to them, so they can react without watching the on-screen notification. Admin-configurable, on by default.
Speech Analytics improvements
PII redaction now covers audio recordings: Call recordings can now be automatically PII-redacted on-device before being uploaded to Speech Analytics, with sensitive segments replaced by a tone, ensuring no PII is stored in recordings. The feature follows your team's existing PII redaction setting in the admin portal.
Agent screen recording for complete interaction visibility: Admins can now enable agent screen recording. It captures desktop activity during calls only and links each recording to its call in Speech Analytics, giving QA teams full context when reviewing interactions.
More consistent transcripts in Speech Analytics: Transcripts, speaker labels, and timestamps are now more accurate and consistent across every call, meaning more reliable reviews for QA teams, and fairer evaluations for agents.
Bug fixes
Voice Translation starts reliably: We've fixed several issues around starting Voice Translation, including when a new call begins while the Krisp mic is already in use, or when the starting prompt is changed.
Refined Deepfake Detection alerts: Shorter whisper tone, clearer messages, and better notification priority.
Correct detected language display: The Widget now shows the parent language as the customer’s detected language, instead of a regional variation (e.g., Spanish instead of Catalan).