Torvald
Knight commander who gives orders that men actually follow
Product
The Audio Drama Maker
This is the product. Everything else on the site is a doorway into it. You bring a script with more than one person in it; CastDub works out who they are, gives each of them a performer, and gives you the controls a director would use.
At a glance
- What it is
- A free AI voice generator and text-to-speech workbench built for scripts: it works out who is speaking, gives every character their own voice, lets you direct the emotion line by line and exports a finished mix — a full-cast audio drama rather than one narrator reading every part.
- What you paste in
- A script, pasted as text. Screenplay format, “Name: line” dialogue and ordinary prose with dialogue tags are all read as written, and so are the em-dash conventions used in Spanish, French, Portuguese and Russian. A note in brackets at the start of a line, or a line that is entirely inside brackets, becomes that line's acting direction: it is performed rather than spoken, and you can edit it in the line list. There is no file import, so paste the text instead of attaching a document.
- What you get back
- One mixed MP3 on every plan. SRT subtitles from Plus upward — written from your script rather than transcribed from the audio, so names are spelled right and the timings already line up. Per-character stems, one full-length track per character plus the mix, come as a ZIP on Pro. No WAV export.
- Cast and voices
- 40 library characters, each with a portrait, a persona and a sample. Behind them sit 10 base voices — 4 female, 5 male, 1 androgynous — and the same 10 serve every language. Inside one project no two characters are given the same base voice until all 10 are taken, so a scene of up to 10 speakers separates cleanly; past that, two characters share a voice and are told apart by persona and by the direction each line carries. You choose the character, not the base voice. You can also clone your own voice by reading a short consent statement aloud — up to 3 per account — and give it to any character; a cloned voice reads in all 9 languages and follows line speed, but does not act out emotions or directions.
- Languages
- 9: English, 简体中文, 日本語, 한국어, Deutsch, Français, Español, Português, Русский. Every character performs in all of them — a character is a persona, not one recorded take — and a project's language is fixed when the project is created. Outside English you get the language's common accent; regional accents cannot be picked and are not promised. We have not had native speakers audition every language, so run your own lines through it before you commit.
- Emotion and direction
- Every line is performed by an AI voice actor working from a written direction, so the 7 emotions (neutral, happy, sad, angry, fear, whisper, excited) are acted rather than faked with settings. A resting emotion per character is available on every plan; overriding emotion line by line is Pro only, and 5-step speed control (0.75×–1.25×) comes with Plus and Pro. Regenerating a line gives a genuinely different take in the same voice, and editing one line re-synthesises only that line.
- Price and open beta
- A credit is one character of script, and credits are spent only when you download — generating and previewing cost nothing, within a ceiling of 30 lines per 24 hours on Free and up to 2,000 on Pro. Free is a one-time pot of 4,500 credits, about 5 minutes of finished audio, with 2 cast members per project. Paid plans are Basic $5, Plus $19, Pro $29 a month, 10% off yearly. Payments are not live yet: during the open beta every account gets the full Pro feature set at no cost, but a commercial licence still comes with a paid plan once plans open.
- What it cannot do yet
- Voice cloning covers your own voice only, and a cloned voice reads the words without acting out emotions or directions. There is no file import, no WAV export, no pronunciation dictionary, no pitch control, and no way to pick a specific base voice or a regional accent. No language has real child voices: child parts are played by adult performers, the way animation has always done it. Characters live inside one project and do not carry over to the next.
Voices to try it on
Morrow
Something under the floorboards that learned your name
Wren
Bedtime-story narrator, slow and warm, never wakes anyone up
Juno
Podcast host who thinks out loud and never reads a script
Kestrel
Low, weary swordswoman who has already lost this war once
Ashford
Velvet-voiced villain who apologises before he ruins you
Hale
Documentary narrator with a hand on the listener's shoulder
Cass
Seventeen and furious about it, in a good way
What it does
A pasted script, no template needed
Screenplay format, Name: prefixed lines, and ordinary prose with dialogue tags are all parsed as written. Paste from a document, a screenwriting app or a spreadsheet column. There is no file upload — .txt, .docx and .srt files are not imported, so copy the text in.
Auto-casting
Speaker detection splits the script by who is talking, then proposes a library character for each speaker based on how the script presents them. It is a proposal, not a decision — every assignment is overridable, and an override persists for the rest of the project.
Line-level emotion
Emotion is a property of the line, not of the character. Set a resting register per character and mark only the lines where it breaks. This is the single mechanism that separates a read-through from a performance, and it is the main thing Pro adds: on the other plans every line follows its character's resting register.
Characters, set once per project
Name, persona, gender, age band, library character and default emotion, set once inside the project. No two characters in a project share a voice until the ten base voices run out, and adding one later does not reshuffle the rest. Characters do not carry between projects, so a series picks the same library characters again in each episode.
Only changed lines regenerate
Edit a line and only that line is synthesised again; everything else keeps the audio it already has, exactly as you last heard it. For a long project this is the difference between revising freely and dreading every rewrite — three changed lines cost three lines of generation, not a full re-render.
Exports that fit a real pipeline
A mixed MP3 on every plan, per-character stems as a ZIP on Pro for scoring and sound design, and an SRT from Plus generated from your script rather than transcribed — so names are spelled correctly and timings already line up.
How to use it
- Paste the script
Copy it in as text. No reformatting step, no template to fill in.
- Check the cast
Review the proposed voice per character, audition alternatives on the same line, and lock in what fits.
- Set resting registers
Give each character a default emotion. Most of the scene runs on the default.
- Direct the turns
Mark the specific lines where the register breaks — usually three to five per scene.
- Render and listen straight through
Pacing problems are invisible line by line and obvious end to end.
- Export
Mix, stems or subtitles, depending on where the audio is going next.
Why casting is the product and synthesis is a component
Speech synthesis is now a commodity. Several providers produce audio that is good enough for drama, and the gap between them is smaller than the gap between a well-cast scene and a badly cast one. What is not commodity is everything around the synthesis: working out who is speaking, keeping a character stable across months, directing individual lines, and getting audio out in a form you can mix.
That is why the provider layer here is an interface with several implementations behind it. If a better engine appears next year, it becomes a configuration change rather than a rebuild, and your projects, casting and line direction carry over unchanged.
What the tool deliberately does not do
It does not write scripts. Generated dialogue performed by generated voices is a product nobody has asked for twice, and the interesting problem in audio drama is production rather than ideas.
It does not mix. No music beds, no reverb, no stereo placement. Those belong in an audio editor where you can hear everything together, and folding them into a rendering tool has reliably produced something worse than either.
It does not overlap dialogue in a single render. Talking over someone is a real dramatic device and it is a mixing operation — export stems and slide them. One drag in any editor, and it is the most realistic thing you can do to an argument.
Where it is honestly weak
Physical vocalisation is the current limit: screaming, sobbing, laughing while talking, grunts of effort. These are breathing events rather than speech events and they are where a listener identifies audio as generated. Write around them where you can, and record the two or three lines that carry them where you cannot. Mixing recorded and generated audio in one production is normal — it happens in your audio editor, since recordings cannot be brought into a project — and the seam does not show when the writing carries the moment.
Very long emotional monologues are the second weak spot, for a related reason: a human performer makes a decision about subtext every few seconds across a monologue, and nothing here attempts that.
Frequently asked questions
What file formats can I import?
None — there is no file upload. Paste the text itself, in screenplay format, Name: lines or prose with dialogue tags. From a Word document or a subtitle file, copy the text out and paste it in.
Does auto-casting always get the speakers right?
In script formats, almost always. In prose it depends on how explicit your attribution is — long unattributed exchanges and characters referred to by epithet are the reliable failure cases. Scanning the assignments before rendering takes a minute.
Can I change one line without re-rendering everything?
Yes, and this is the feature that makes long projects possible. Change a line, re-render that line, everything else keeps its existing audio.
Do characters carry across projects?
No. A character exists only inside its project. For a series, create a new project per episode, pick the same library characters and write the same personas. The same voice is not guaranteed across projects, because voices are kept distinct in the order characters are added within each project — so listen to your lead's first line before you export.
Can I add music and sound effects?
Not here. Export the stems and score them in an audio editor. This is a deliberate boundary, not a missing feature.
Which plan gives me line-level emotion control?
Pro. On Free, Basic and Plus every line follows its character's default emotion, which is good enough to check that a scene works but not to place a specific turn on a specific line.
Cast your first scene tonight
Paste a script, let CastDub assign a voice to every character, and export a finished drama.
Create for freeNo credit card. Free gives you a one-time pot of credits.