
Midnight Radio
Velvet · narrator
● Main narrator · urban legends
Low, warm and steady, like late-night radio in your ear. The default for urban legends and thriller recaps; switch to whisper on the key lines.
A horror story voiceover is half voice and half timing: the narrator talks close to your ear, the characters get more frightened with every line, and the scariest sentence is the one said most calmly. With AI Drama Voice you paste your creepypasta or horror script, the AI separates the narrator from each character and casts a voice for every one, then performs each line with the direction you wrote in brackets — whisper, trembling, terrified. You download an MP3 and a line-by-line SRT, and add music and sound effects in your editor.
Updated Nov 5, 2026Fact-checked Nov 5, 2026
Our sample · Poisoned Wine in the Cold Palace
Chinese · Script excerpt · 4 characters

NarratorCalm
冷宫的门三年没开过。今夜,贵妃端着一壶酒走了进来。
The cold palace door had not opened in three years. Tonight, the Noble Consort walked in carrying a jug of wine.

Noble ConsortSarcastic
姐姐,这是陛下赏你的,喝了吧。
A gift from His Majesty, sister. Drink up.

Deposed consort · Shen ZhiweiCold laugh
你的孩子,真的在你肚子里吗?
And your baby, is it really in there?

EmperorFurious
都给朕住口!
Silence, all of you!
The studio opens on the sample Snow in the Cold Palace: three narration lines go to the late-night radio voice, the consort's cold laugh and taunts go to the villain voice. Replace it with your own story in the format Name (whisper): line and watch the AI split the roles — this step costs no credits.
Narrator
Midnight Radio · 4 lines
3 roles · 7 lines
561 / 8,000 charsSame cast · Chinese
The first clip is our own sample, Poisoned Wine in the Cold Palace: palace intrigue in Chinese, with a narrator who keeps his voice down and a villain who smiles while she threatens; the English meaning sits under each line. The two clips after it are reference recordings for close-to-the-ear narration and a ghost who sounds sweet — they were not made with our voices.
Script excerpt · Our sample · Poisoned Wine in the Cold Palace

NarratorCalm
冷宫的门三年没开过。今夜,贵妃端着一壶酒走了进来。
The cold palace door had not opened in three years. Tonight, the Noble Consort walked in carrying a jug of wine.

Noble ConsortSarcastic
姐姐,这是陛下赏你的,喝了吧。
A gift from His Majesty, sister. Drink up.

Deposed consort · Shen ZhiweiCold laugh
你的孩子,真的在你肚子里吗?
And your baby, is it really in there?

EmperorFurious
都给朕住口!
Silence, all of you!
Script excerpt · Rule-based horror · Night-Reading Rules for Booth Four
The voice reading alongWhisper
她读得很稳,像是早就背过这一页。
She reads so steadily, as if she memorised this page long ago.
Narrator
许棠后背一凉。她忍不住抬头,看向玻璃。
A chill ran down Xu Tang's back. She couldn't help looking up at the glass.
Xu Tang
周老师,您听见了吗?耳机里,还有一个人在跟着我读。
Mr. Zhou, can you hear that? There's someone else in my headphones, reading along with me.
Mr. ZhouCalm
没有啊,许老师。耳机里只有你一个。别停,接着读。
No, Ms. Xu. It's only you in there. Don't stop. Keep reading.
Script excerpt · Supernatural · The Fourteenth Floor That Isn't There
He Chuan
我叫何……
My name is He—
Old ZhouFurious
别告诉她!
Don't tell her!
Narrator
何川一口咬住了舌头,嘴里泛起一股铁锈味。
He Chuan bit down on his tongue and tasted rust.
The woman from the 14th floorTender
何什么呀?没关系,我会知道的。
He what? Never mind. I'll find out.
The woman from the 14th floor
你们都是这样。那天晚上,我加完班,按了电梯,一直等,一直等。电梯来过三次,门开了,里面的人看了我一眼,又把门关上了。
You're all the same. That night I finished late and called the lift, and I waited and waited. It came three times. The doors opened, the people inside looked at me, and closed the doors again.
Most of the work in a horror story voiceover happens in the script: where to break lines, which line drops to a whisper, where everything stops. Do these five things, then let the AI perform it.
Step 1
Break the text into short rows, one image or one action each, so a line like 'The light went out.' stands alone. Long sentences sound read; short ones sound like someone holding their breath.
Tip · The export leaves a short pause between rows, so shorter rows mean a slower pace.
Step 2
Use a single voice for all narration, mark it calm, and switch to whisper on only two or three key lines. The steadier the narrator, the louder the characters' panic.
Step 3
Start every line of dialogue with the character's name and put fear in brackets: Lin Xi (trembling): Who's out there? Rows without a name go to the narrator.
Step 4
Hand the most frightening sentence to the person who should not be calm, and have them say it calmly: a ghost in a gentle voice, a landlord who sounds bored. That contrast chills more than a scream.
Step 5
After generation, listen to each line, change the bracketed note where it misses, and redo only that line. Download the MP3 and SRT, then lay drones, heartbeats and door slams on the subtitle timestamps in your editor.
These are the voices we reach for most in horror and thrillers. The Chinese persona voices speak all nine languages; the native English voice speaks English only.

Velvet · narrator
● Main narrator · urban legends
Low, warm and steady, like late-night radio in your ear. The default for urban legends and thriller recaps; switch to whisper on the key lines.

Classic · folk tale
● Storyteller · folk tales
A traditional storyteller's cadence, made for 'that winter, the old tree by the village gate' style folk horror and ghost tales.

Soft · soothing
● The ghost who sounds kind
A soft, soothing female voice saying terrible things gives the biggest contrast. Use tender or whisper for the woman on the fourteenth floor.

Cold · villainous
● Cold ghost · keeper of the rules
Cool and distant; her cold laugh and calm both press down on the listener. Good for ghosts, sinister matrons and whoever is behind it all.

Crisp · broadcast
● Rule announcements
A precise broadcaster reading 'Rule one: do not answer the knock'. The more correct she sounds, the creepier it gets.

Clear · energetic
● The panicking young lead
A clear young male voice that lets you hear bravado turn into fear: the new night guard, the student who walks into the haunted flat.

Warm · grounded
● The one who knows
Deep and weighty, best for a lowered-voice warning: the old security guard, the landlord, the village elder. 'Don't look back' is all he needs.

Mature · poised
● The calm investigator
Mature and composed, for the detective, the psychologist, or the only person in the story who sees the truth.

Over-the-top · comic
● Comedy horror only
An over-the-top cartoon villain. Use it for parody and prank 'horror', never for a story meant to scare.

American · deep
● English horror narrator
A deep American male voice, English only, for creepypasta readings and late-night story channels.
A lot of people's first AI horror voice attempt is to drop the whole story into a text-to-speech reader. It comes back sounding like the news: every sentence equally fast and equally loud, and the scary part flies past. A horror story voiceover really comes down to four controls: volume and distance, breath, pace and silence. The table shows how to write each one in an AI Drama Voice script.
| Delivery | How to write it | Where it belongs | Example |
|---|---|---|---|
| Whisper | Write (whisper); softly, quietly and hushed are read as whisper too | Close-to-the-ear narration, the ghost's lines, the key clue | Narrator (whisper): Every night at 3 a.m., it switches on by itself. |
| Breathy / trembling | Write (trembling), (breathy) or (gasps); words outside the eight tags are passed on as a delivery note | A frightened lead, someone pretending to be fine | Lin Xi (trembling): ...Who's out there? |
| Slow | Short sentences, more rows, full stops and ellipses | Build-up: opening a door, climbing the stairs | The light went out. / A note slid under the door. |
| Sudden stop | Cut a sentence off with a dash, then start the next row in a completely different tone | The beat before a twist, a call that drops | But I live on the— / Landlord (calm): Nobody lives on four. |
Whisper, calm, tender and cold laugh are ready-made options among AI Drama Voice's eight emotion tags. Trembling, terrified and breathy are not tags, but as long as they sit in brackets and stay within 20 characters, the system sends them with that line as a delivery note. So Lin Xi (terrified): But I live on the fourth floor is a valid line.
Here is a short script we wrote for this page; paste it into the studio to try it. Notice that the narrator stays calm almost the whole way, with one whisper. The lead is the one who is frightened, and the person with the scariest line is the calmest of all.
| Speaker | Emotion / note | Line |
|---|---|---|
| Narrator | calm | My building is old. The stairwell light only comes on when it hears a sound. |
| Narrator | whisper | On my third night there, I noticed it switches on by itself, every night, at 3 a.m. |
| Lin Xi | trembling (note) | ...Who's out there? |
| Narrator | none | No answer. The light went out. Slowly, a folded note slid under the door. |
| Lin Xi | whisper | It says... don't turn on the light. I'm behind you. |
| Landlord, on the phone | calm | Oh, Xiao Lin, I forgot to mention. Nobody has lived on the fourth floor for years. |
| Lin Xi | terrified (note) | But... I live on the fourth floor. |
In a horror story the narrator leads you in and the characters are the ones getting scared. With one voice for both, listeners lose track of who is speaking and the tension leaks away. AI Drama Voice detects the narrator and every character separately; the narrator starts on the late-night radio voice and the characters are cast by gender, age and role, and you can swap any of them in one click. A thriller recap is almost all narration, which makes it even more important to put quoted dialogue on its own row with a name, so 'she said' switches to a different voice.
About the 'female ghost' voice
AI Drama Voice has no dedicated ghost voice and no voice changer, echo or reverb effects. The usual approach is the soothing female voice or the cold queen voice with whisper or a note like (breathy) or (drawn out), so she says frightening things gently or flatly. Add the hollow, echoing quality with reverb in your editor.
AI Drama Voice generates voices only: the MP3 has no score, ambience or effects. The other half of a horror story voiceover is sound design — a low drone, a heartbeat, dripping water, a hinge, electrical hum. The SRT gives every line its own timestamp, so you can drop a power-cut effect exactly where 'The light went out' starts and leave silence right before 'I'm behind you'. Our guide to syncing SRT files in CapCut shows how to line subtitles up with audio.

Horror and gore usually face stricter review on short-video and audio platforms, and gory thumbnails, titles or captions can limit how far a video travels, so read the community guidelines of the platform you post on. Work voiced with AI may also need an AI-generated content label; see how to add an AIGC label.
Both are suspense, but a crime story moves on cool, clipped dialogue while English horror leans on the silences in the narration. Both are reference clips, not made with our voices.
Script excerpt · Crime thriller · The Cuckoo Called Twelve Times
Zhan WangCalm
邵队,你数得真准。那你还记不记得,十年前,也有一个人跟你说过,他听见那只钟叫了十二声?
You counted perfectly, Captain Shao. So do you remember, ten years ago, someone else told you he heard that clock call twelve times?
Narrator
邵青握笔的手停住了。
Shao Qing's pen stopped moving.
Shao Qing
老街金店案。詹朗。
The Old Street jewellery case. Zhan Lang.
Zhan Wang
我哥。
My brother.
Xiao Ge
邵队,那个案子,不是早就结了吗?
Captain, wasn't that case closed years ago?
Shao Qing
结了。那年我二十六岁,刚进刑警队。那是我破的第一个大案,那份口供,我到今天还背得出来。
It was. I was twenty-six, new to the squad. It was my first big case. I can still recite that confession.
Script excerpt · English horror · It Knocks in My Brother's Voice
Priya
Okay. No. Theo, tell me that's a recording.
Theo
It's some guy from town doing impressions. Hey, buddy! Great Owen! Now go home!
Owen
Theo, you still owe me twenty bucks from the gas station. And you told Priya you quit smoking. You didn't. Mags, please.
NarratorCalm
Theo stopped laughing. Nobody else had been at the gas station. Nobody else had seen the cigarette.
MayaWhisper
Ask it something, Owen. Something only you would know.
Lin Xi (backing away slowly, shaking, barely audible): is too long to be read as a note, so the whole row becomes narration and the narrator reads out 'Lin Xi' too. Two or three words are enough.
Emotions are matched by keyword. (quietly furious) contains 'quietly' and becomes a whisper; (laughing, crying) contains 'crying' and becomes choked. If the word on the page isn't the effect you want, rephrase it, for example (bitter smile).
Rows labelled with any of these names go to the narrator voice. If you want the lead's inner voice in the lead's own voice, write Lin Xi (whisper): ... instead of VO: ...
A row that only says (the door creaks) is read aloud by the narrator; it never becomes a creaking door. Keep sound cues in a separate cue sheet and add them in post.
If every line is hushed, the ear gets used to it within a minute. Save the whisper for a few key lines and surround them with calm narration, so the voice feels like it suddenly leans in.
Translated versions keep each line's original emotion tag. English horror favours short, plain statements, so glance at the translation and shorten long lines in the studio before you generate.

Daily scary stories with one voice for the narrator, one for the rule announcements and one for the victim, voiced from one script in one pass.

Recaps are mostly narration. Put the characters' own words on separate rows and their voices cut into the recap. For more on recap scripts, see short-drama recap voiceover.
Multi-character supernatural and crime episodes, each line tagged, exported as MP3 + SRT to line up with the picture before music and effects go on.
Tick English on a Chinese ghost story and the same cast voices it in English, or swap the narrator for a native English voice.
Roughly three kinds: general text-to-speech readers, the voiceover tools built into video editors, and script-based tools that cast a voice per character. AI Drama Voice is the third kind: the narrator and every character get their own voice, and each line can carry a direction like whisper or trembling, which suits multi-character horror stories and thriller shorts.
Fix the script before the voice. Break long sentences into short rows, keep the narrator calm and mark only two or three key lines (whisper), and give the scariest line to the calmest character. Listen line by line after generating and redo only the lines that miss. The five steps above walk through it.
There is no dedicated ghost voice or voice-changer effect. The common approach is the soothing female voice or the cold queen voice with whisper or a note like (breathy) or (drawn out), so she says frightening things gently. Add the hollow, echoing quality with reverb in your editor.
Usually yes, and you add it yourself. AI Drama Voice generates voices only; the MP3 has no music or effects. The SRT has a timestamp for every line, so you can place drones, heartbeats and door slams exactly, and a moment of silence before the key line often works better than wall-to-wall music.
Most thriller recaps use a low, steady male voice or a composed female voice, such as the late-night radio narrator or the mature heroine voice; folk horror suits the traditional storyteller. Preview them in the voice library, then swap with one click in the studio.
Yes. Tick English and each line is rewritten in English, then voiced by the same cast with its original emotion; you can also switch the narrator to a native English voice. See English dubbing.






Paste the story, split the narrator from the cast, write whisper and trembling in brackets — then take the MP3 and subtitles and add your own sound design.
Billed by characters generated · Failed runs refunded automatically · Alipay, WeChat Pay and cards