Gemini Notebook (formerly NotebookLM)‘s Audio Overview turns uploaded documents into a two-person conversational Podcast. The English version is one of the stronger AI tools available today: the two hosts sound close to real people. The Chinese version’s accent and pauses have not reached the same level, but it is still practical for quickly “listening through” a long report.

Google renamed NotebookLM to Gemini Notebook on July 16, 2026. It is the same product, and existing notebooks and links are not affected.

What Is Audio Overview?

Google launched this feature in September 2024, and it became one of Gemini Notebook’s most popular features. The system reads all uploaded sources and generates a 10-20 minute audio file where two AI hosts discuss the key points in conversation.

It is different from simple text-to-speech. It digests the material, reorganizes the points, and explains the core concepts conversationally. The hosts add to each other’s points and sometimes ask follow-up questions. This sense of interaction is hard for other AI tools to match.

Audio Overview settings screen

How to Generate a Chinese Podcast with Gemini Notebook

Step 1: Create a Notebook and Upload Sources

Go to Gemini Notebook.google, create a new Notebook, and upload the material you want to turn into a Podcast. It supports PDFs, Google Docs, web links, YouTube videos, plain text notes, and audio files.

A Notebook can hold multiple sources, and the AI synthesizes all of them to generate content. Use 3-5 related documents. If there are too many, the AI can lose focus.

Step 2: Open the Studio Panel

There is a Studio section on the right. Open it to see the Audio Overview option.

Before generation, you can enter instructions telling the AI what Podcast style you want. For example:

  • “Explain in simple terms for an audience without technical background”
  • “Focus on the experiment results in Chapter 3 and only briefly cover the other chapters”
  • “Use a lighter tone, like friends chatting”

You can skip this step, and the AI will decide on its own. With instructions, the output is noticeably more focused.

Step 4: Generate and Wait 2-5 Minutes

Generation speed depends on source volume. A 10-page PDF takes about 2 minutes. Five documents together may take 5 minutes. You will receive a notification after generation completes.

Step 5: Listen and Download

The generated audio can be played directly in the browser and downloaded as an MP3. There is no watermark, and it can be used freely.

Chinese vs English: Where the Quality Gap Is

Voice quality is the most important thing to mention about this feature.

English version: highly polished. The two hosts have expressive intonation, respond to each other, and occasionally make jokes. Community feedback generally places it among the best AI Podcast generation tools available.

Chinese version: listenable, but not yet natural. Specific issues include a mechanical accent, awkward pause placement (no pause where one is needed, forced breaks where one is not), and flatter intonation than English. It sounds more like two people reading a script than chatting.

But the content extraction ability is the same. The Chinese version is just as capable as the English version at finding key points and organizing arguments. The issue is purely voice synthesis quality.

Practical choice: if you want to quickly “listen through” a document yourself, Chinese is good enough. If you want to share with others or publish publicly, use the English version, or use the transcript feature to get text and record it yourself.

Audio Overview output screen

Real Use Cases: How to Use Audio Overview

Scenario 1: Quickly Digesting Long Research Reports

When a 40+ page industry report arrives and the deadline is tight, reading the whole thing is unrealistic. Put the PDF into Gemini Notebook and generate a 15-minute Audio Overview. Listen during commute, and you can catch the report’s core arguments and key data points, enough to speak in the meeting.

Scenario 2: Quality Checking Your Own Article

After writing a long article, put the draft in and generate a Podcast. Listen to how the AI explains the content. If a section sounds circular or logically rough, the original text usually has the same issue. Catching problems by ear is often easier than rereading the piece over and over.

Scenario 3: Turning Reading Notes into Shareable Content

After finishing a book, add your notes plus a few related reviews. The AI synthesizes them into a structured discussion, which has higher information density for listeners than simple narration.

Scenario 4: Organizing YouTube Transcripts First, Then Generating a Podcast

To turn several YouTube interviews on the same topic into a Chinese Podcast recap, use this flow:

  1. Find 3-5 YouTube videos on the same topic, such as videos discussing the same new model
  2. Paste their URLs directly into the same Notebook’s Sources, and Gemini Notebook automatically grabs subtitles for all videos
  3. Do not generate Audio Overview yet. First ask Gemini Notebook to output each video’s chapter outline and key quotes, then organize them into one summary document (called Briefing Doc in the Gemini Notebook interface)
  4. Add this summary document back into the Notebook as a new note, and remove the original video sources to keep the focus clean
  5. Only then click the Audio Overview generate button

The reason for the extra steps: if video subtitles go directly into Audio Overview, the two hosts will jump around and touch a little of every video. Converging first into one summary makes the generated Podcast much more focused.

The YouTube link feature is smoothest under a Google AI Pro subscription (Taiwan NT$650/month, includes Gemini Notebook Pro quota). For detailed steps, see the Gemini Notebook Transcript Tutorial.

How to Download the Gemini Notebook Podcast MP3 File

After generation finishes, the Audio Overview block has a download button in the upper right. Click it to save an MP3. The file has no watermark and no usage restriction; it can be used in your own Podcast channel, shared with colleagues, or kept as study material.

File size depends on length. A 15-minute audio file is usually about 10-15 MB.

Gemini Notebook Podcast vs TTS Tools Such as ElevenLabs

Other AI Podcast generation tools such as ElevenLabs and Wondercraft have a different positioning. Those tools need a finished transcript first; the AI handles voiceover. Gemini Notebook’s Audio Overview directly generates both dialog content and voice from the original materials. The whole process only requires uploading documents.

In other words, other tools “read out a script that is already written,” while Gemini Notebook “digests the material first and then explains it to you.”

Audio Overview currently does not support custom voices, background music, or segment-level editing. To produce a full Podcast show, professional tools are still needed. But for quickly turning documents into listenable content, Gemini Notebook is the lowest-effort option.

Pitfall Notes

Too Many Sources Make Quality Worse

Put more than 10 documents in at once, and the AI tends to jump from source to source, touching each one but going deep on none. Keep it within 3-5 sources, and quality is much steadier.

The Free Tier Only Has 3 Generations Per Day

Before using the quota, make sure your sources and instructions are set. Do not waste attempts on testing. To raise the quota, upgrade to the corresponding Google AI plan: Plus 6 per day, Pro 20 per day, Ultra 200 per day.

Chinese Can Occasionally Mix in Simplified Terms

Even when uploading Traditional Chinese material, the generated Podcast may occasionally use Simplified Chinese terms, such as “視頻,” “資訊,” or “使用者.” It does not prevent understanding, but it can feel awkward for a Taiwan audience. At the moment, instructions cannot fully prevent this.

Penchan’s Take

Penchan mainly uses Gemini Notebook for transcript output, then sends the transcript to another large model for follow-up analysis. Because the current Chinese voice quality of Audio Overview is not yet strong enough for public channels, this feature is not part of Penchan’s daily Podcast production flow. The steps and scenario judgments in this article are based mainly on official documentation and community testing.

The overall observation on Gemini Notebook’s Chinese support: text output quality is high, while Chinese text inside image and slide generation is severely distorted and is a known pitfall. Voice sits between the two: usable, but with a ceiling.

If you want to turn material into listenable content and do not mind quality being clearly behind English, this tool is efficient for “quickly absorbing long text.” For formal Podcast production, the practical route is still the traditional one: write the script yourself and find voiceover.

Further Reading

This article compares AI tool features and subscription plans. It does not constitute securities or investment advice. Actual pricing should follow the latest official announcements from each platform, and the information here may become outdated.

— Penchan