[Voice Chat] Speaking pace becomes too fast during long explanations; add persistent speed controls

What variant of Codex are you using?

Codex/Chatgpt App (Desktop) — Voice Chat

What feature would you like to see?

Please add a manual and persistent speaking-speed control for Voice Chat.

During long or information-dense explanations, the assistant often begins speaking progressively faster. This is especially noticeable during technical explanations, code walkthroughs, lectures, and other responses that contain a large amount of information.

I can ask the assistant to “speak more slowly,” and it will usually slow down temporarily. However, after a short period of time, the speaking pace gradually becomes fast again. Repeatedly interrupting the conversation to ask it to slow down makes Voice Chat much less useful for learning and technical discussions.

A user-controlled speed setting would be more reliable than controlling the pace through natural-language instructions.

Possible implementation:

• Add a speaking-speed selector or slider to the Voice Chat interface.
• Provide presets such as 0.75x, 1.0x, 1.25x, and 1.5x.
• Keep the selected speed active for the entire Voice Chat session.
• Ideally, remember the user's preferred speed across future sessions.
• Allow the speed to be adjusted during an active conversation.
• Apply the setting consistently to both short answers and long explanations.
• Change only the audio delivery rate, without reducing the amount of detail or changing the model's reasoning.

Expected behavior:

Once a user selects a slower speaking rate, Voice Chat should consistently maintain that rate until the user manually changes it. The speaking pace should not gradually accelerate during longer responses or conversations.

Additional information

No response

Start at the Codex/ChatGPT desktop Voice Chat interface and inspect how audio settings are handled during an active session. Define completion as a user-selectable speaking rate that remains consistent through short and long responses, can be adjusted during a conversation, and is optionally retained across sessions.