NotebookLM Audio Overview: how it works and how to use your own voices
NotebookLM turns your notes, PDFs and links into a podcast-style conversation between two AI hosts. It is a quick way to listen to material instead of reading it, and many people then want the same format for their own channel, in their own voice.
This guide explains, according to Google’s NotebookLM help, how Audio Overviews work and what you can change, and how to voice a two-host episode yourself with SuperClonica when you want to choose the voices or edit every line.
What an Audio Overview is
Google describes Audio Overviews as deep-dive discussions between AI hosts that summarize the key topics of the sources in your notebook. Open or create a notebook, add sources and choose Audio Overview in the Studio panel; it is generated in the background while you keep working. Google notes that the overviews, including the voices, are AI-generated and may contain inaccuracies or audio glitches.
Formats, language and length
You can choose Deep Dive, the default conversation; The Brief, a single-speaker summary under two minutes; The Critique, constructive feedback on material such as an essay; or The Debate, a formal back-and-forth between perspectives. More than 80 output languages are supported, the length can be Shorter, Default or Longer, where Longer is English only, and you can add a prompt to focus the hosts on specific topics.
Interactive mode, sharing and download
In English, newly generated overviews have an interactive mode: press Join, and when the hosts call on you, ask a question by voice. You can share a link, share the notebook or download the audio as a file; downloads can be blocked for notebooks that use Play Books sources. Playback speed is in the More menu.
Your own two-host episode
The help describes choosing the format, language, length and prompt, not the hosts’ voices. When you want specific voices, your own voice or full control of the words, write or edit the conversation yourself as a table with two columns, voice and line, one row per line. Then install the free SuperClonica engine with python -m pip install superclonica-cpu and, in Python, after import csv, from superclonica import Superclonica and tts = Superclonica(), then for i, row in enumerate(csv.DictReader(open("episode.csv", encoding="utf-8"))): audio, rate = tts.synthesize(row["line"], voice=row["voice"], lang="en", seed=3); tts.save(f"line-{i:03}.wav", audio). Join the files in order in any audio editor, with a short pause between speakers.
Choosing the two voices
Pick two voices that are clearly different, for example a lower male narrator and a brighter female one: hear catalog voices free on their pages with 30 to 100 characters, no account needed. For a show in your own voice, train it in the studio from a clean 5 to 10 second recording; training on your own NVIDIA graphics card is free. The engine runs offline in 31 languages with no character limit.