Narration Voices
Anything Scholarly speaks — a video lecture, a podcast, a document read aloud as an MP3, a voice note in chat — is narrated by a voice you can choose. The picker lists the exact voices that will speak, by name, and every one has a preview, so you hear the narrator before you spend a credit.
Picking a Voice
The Voice row sits on the customize step of every creator that narrates. Open it and you get:
- Auto — Scholarly picks a voice that suits the material. If you have no opinion, this is the right answer.
- Every named voice, each with a play button for a short demo.
- Search, which matches on the voice's name, its description, and what it is good for — so "documentary", "calm", or "training" all find something, not just a name you already know.
The choice is made per creation, not once for your account, so a training module and a late-night revision podcast can sound completely different. Your last pick is remembered as the starting point next time.
The Voices
Which voices you see depends on the language the narration will be in. You do not have to do anything for this — the picker lists whichever set will actually speak your creation.
English
English narration uses a set of ten voices, five female and five male:
| Voice | Voice | ||
|---|---|---|---|
| Nadine | Female | Bram | Male |
| Tabitha | Female | Gareth | Male |
| Noelle | Female | Lucian | Male |
| Meredith | Female | Weston | Male |
| Sabrina | Female | Vincent | Male |
They are a small set on purpose: ten voices you can actually listen through in a minute, rather than a wall of names.
Every other language
Narration in any other language is spoken by our multilingual voices — a larger set that reads over 70 languages natively, each with its own character and a short description of what it suits. A few examples of how they differ:
- Rasalgethi — informative and measured, for history lectures and audiobooks.
- Erinome — clear, articulate, and precise, for training walkthroughs and instructions.
- Despina — smooth and polished, for professional e-learning.
- Achernar — soft, calm, and soothing, for quiet narration.
- Puck — upbeat and energetic, for podcasts and reviews.
- Gacrux — mature and authoritative, for serious presentations.
The full list is in the picker, with a preview on each. Search is the fast way through it: type the tone or the job rather than scrolling.
How the Language Is Decided
The language of the narration decides which set of voices speaks it, and it is decided once, before any audio is recorded, so a single lecture or episode never changes narrator part-way through.
- You picked a language — that is the language, and the picker shows the voices for it.
- You left it on Auto — Scholarly reads your source and narrates in its language. When the material is too short or too mixed to call confidently, it uses the multilingual voices, which read everything.
If you choose an English voice and the creation turns out to be in another language, it is narrated by the matching voice from the multilingual set rather than being read badly in the wrong accent. Right-to-left languages such as Arabic and Hebrew are narrated and laid out right to left. See Languages in Scholarly.
Where the Voice Picker Appears
| Where | What it narrates |
|---|---|
| Video Lectures → Customize → Voice | The whole lecture, chapter by chapter |
| Podcasts → Customize → Voice | The hosts and guests of the episode |
| Text-to-Speech → Customize → Voice | A document read start to finish as one MP3 |
| Settings → Features | The assistant voice used for chat voice notes and read-aloud |
Voice Mode, the live two-way conversation, has its own separate set of conversational voices — it is a different job, so it is a different list.
Speed
Where a creation offers a Speed control — Text-to-Speech, for example, at 0.75×, 1×, 1.25×, or 1.5× — the speed is recorded into the audio rather than applied at playback, so a faster narration really is a shorter file, and it sounds exactly as fast as you asked.
Read-aloud elsewhere in Scholarly starts speaking almost immediately and holds a steady volume across one piece of audio and the next, so a long document does not stutter between sections.
Telling Us What You Think of a Voice
Finished lectures and podcasts carry a thumbs-up / thumbs-down control. That is how we learn which voices land, and it is worth using when a narrator is either excellent or clearly wrong for the material.
We do occasionally narrate new creations with a different voice while we work out which ones sound best. Anything you already made keeps the audio it was made with — a lecture never quietly changes voice after the fact.
Tips
- Preview two before you commit. A voice that reads a product spec well is often wrong for a story book, and the demos take five seconds each.
- Search the job, not the name. "training", "documentary", "calm", "announcement" all find voices; nobody remembers the names cold.
- Leave it on Auto for a first pass, then pick deliberately for anything you will hand to other people.
- Set the language explicitly for anything client-facing. Auto is good, but a pinned language is certain.
- A new voice means a new creation. Re-narrating the same document with a different voice is a fresh creation and is counted as one.
Related
- AI Video Lectures — where the voice picker matters most.
- AI Podcasts — multi-speaker casting.
- Text-to-Speech — a document read aloud, with speed and cost quoted up front.
- Voice Notes in Chat — spoken replies in a conversation.
- Voice Mode — the separate live conversation voices.
- Languages in Scholarly — interface, AI output, and spoken narration.