NotebookLM Audio Voices: What You Can Customize and What Is Not Listed
Updated November 7, 2026
Direct answer: The documented Audio Overview controls in Google’s current Gemini Notebook help let you customize the format, language, length, and focus. There is no documented voice picker for choosing a named voice. Google’s current Help materials do not establish that voice customization is universally impossible; they simply do not list such controls.
Documented Audio Overview controls
The Audio Overview panel lists:
- Deep Dive: two hosts.
- Brief: one speaker.
- Critique: two hosts.
- Debate: two hosts.
- Language: choose an output language offered in the panel.
- Length: Shorter, Default, or Longer; Google currently marks these length options as English-only.
- Focus prompt: a way to direct the discussion toward a topic or area of expertise.
This is format-level direction rather than voice casting. Choosing Brief means using the documented one-speaker format; it does not mean choosing a named presenter. Likewise, the two-host formats do not come with documented controls for selecting individual host identities.
What is not currently documented
The current Help materials do not list a voice preset library, age or gender selectors, voice cloning, tone or pacing generation sliders, background sound, or an “Instant Audio” mode.
That distinction matters when writing setup instructions or promising features to staff. It would be inaccurate to describe a “voice dropdown,” voice categories, or a cloning workflow as available. The safer wording is “not currently documented,” because an absence from the documentation is not proof of a universal technical impossibility.
A practical source-to-audio workflow
- Upload sources to a notebook. For a returns-policy lesson, limit the source material to the supplied policy rather than adding assumptions or outside guidance.
- Open the Audio Overview panel. Choose the format according to the lesson’s needs: Brief for one speaker, or one of the two-host formats when a structured exchange is appropriate.
- Set the available options. Select the listed language, length, and focus prompt. Use the focus field to identify the audience, policy topic, and any uncertainties that should be preserved.
- Generate the overview. Treat this as an AI-generated discussion based on the sources, not as a guaranteed word-for-word reading of a script.
- Listen and verify the terms. Generated audio may contain errors or glitches. Compare return conditions, exceptions, and deadlines with the supplied policy, and remove any draft that introduces unsupported details.
- Adjust playback, not identity. If the pace is unsuitable, use More > Change playback speed. That changes the listening experience rather than regenerating the voice identity. Also account for the documented mobile limitations when planning where staff will listen.
Five-line focus prompt for new-staff training
Use the following as a proposed focus prompt:
Create a training Audio Overview for new staff on product returns.
Use only the supplied returns policy; do not add eligibility rules, fees, exceptions, approvals, or deadlines that the policy does not state.
Define the audience as staff who have not previously completed training on this returns policy, and state that definition in the audio.
If the supplied policy does not give a clear return deadline, say that the deadline is not stated clearly and must be confirmed before use.
End by reminding staff to verify each return instruction against the supplied policy before acting.
Three distinctions to keep clear
| Item | What it means | What it does not mean |
|---|---|---|
| Brief format | The documented format uses one speaker. | It does not provide a way to select a named voice. |
| Change playback speed | The generated audio is played at a different speed. | It does not create a new voice identity or regenerate tone and pacing. |
| Your own voice in an interactive exchange | The request concerns who is speaking during that interaction. | It does not demonstrate that your voice has been cloned for generated Audio Overview audio; voice cloning is not listed. |
Before using the training audio
- Choose the format based on speaker count and discussion structure, not on an imagined presenter.
- Keep the training source set limited to the supplied policy.
- Compare every substantive return instruction and deadline with that policy.
- Reject narration that introduces unsupported rules, even if it sounds polished.
- Check the intended listening device while accounting for mobile limitations.
Continue with the next task
Sources and scope
Documentation checked November 7, 2026. Rewarx is the publisher. Covers are MiniMax-generated editorial concepts. Worked examples and diagrams are instructional illustrations, not measured products, tool output comparisons or customer results. Procedures are documentation-based; no live shopping transaction or hands-on software benchmark is claimed.