v7.43

Release Notes

This update adds expanded Voice Mode configuration controls.

Voice Mode

New Features

• Configure Voice Mode assistants with expanded Advanced Options for speech detection, microphone processing, spoken voice, and response speed.
• Choose between Semantic VAD and Server VAD to control how Voice Mode detects when a user has finished speaking.
• Use Semantic VAD eagerness controls to tune how long the assistant should wait for the user to continue speaking before responding.
• Fine-tune Server VAD settings, including activation threshold, prefix padding, silence duration, and idle timeout.
• Choose and preview the realtime voice used by a Voice Mode assistant.
• Adjust spoken response speed for Voice Mode assistants, with 1.0 as the default speed.
• Configure microphone noise reduction with Far field, Near field, or Off options.

Notes

• New Assistants will default to somewhat different settings than the configuration setup previously enforced by the server:
  • Voice: Marin. OpenAI recommends using Marin or Cedar for the highest audio quality.
  • Speed: 1.0x. Previously 1.15x.
  • Automatic turn detection: Semantic VAD
  • Eagerness: Auto, currently equivalent to medium. Previously high.
  • Noise reduction: Far field
  • Timing for Server (Normal) VAD: 0.5 threshold, 300ms prefix padding, 500ms silence duration, no idle timeout. These options were unavailable in previous versions because PingPong only supported Semantic VAD.

Deployment Information

Schema Upgrade	Migration Script	Permissions Update	Task Definition Update	Configuration Update
YES	No	No	No	No

Deployment Details

• Schema Upgrade: New realtime configuration columns in assistants table.

Related PRs

• feat(voice): add configuration options by @ekassos in #1701

Full Changelog: v1289+srv654.web482...v1291+srv655.web483