Text to Speech for Writers: Hear Your Chapter Before You Post

You know who says every line in your chapter, because you wrote it. Your readers only have the page. If two characters share "she" for six lines, a reader can lose track. The same goes for a reply right after another character's action beat (a bit of action, like "She set down her cup."). Rereading your own draft won't show you where this happens.

Hearing it can. LyricWinter finds the narrator and every character who speaks, gives each one their own voice, and reads the chapter back. When a line comes out in the wrong voice, you hear the problem a reader would have felt. Your draft isn't public until you choose to share it.

You do all of this in LyricWinter Studio, the LyricWinter app that turns stories into audio. It works on a fic chapter for AO3, a serial (a story posted in parts) on SpaceBattles or Royal Road, a Substack story, or a novel draft. A Studio project holds a whole story, with one section per chapter. If you read fic rather than write it, see how to listen to fanfiction. Below is a scene written to be hard to follow. Listen before you read on.

Listen first: who says each line?

"Fog on the Marigold" is an original scene written for this blog. Near the end, Captain Ines Varga and Mireille stand at the ship's rail and share "she" and "her" for several lines. Two lines are hard on purpose: "It might.", which has no tag, and the final "Thank you.", which comes right after "She said it so quietly". Listen, note which voice says each one, then open the transcript and decide whether a reader would know.

Every voice here runs on ElevenLabs v4, one of Studio's voice engines (the AI that speaks the lines).

Here is what Studio did. It gave both hard lines, "It might." and "Thank you.", to Captain Ines Varga. Both are Ines's lines, so Studio got them right. Every other line at the rail went to the right woman too, including "That was reckless," she said. If you heard it the same way, the page is clear. If you hesitated, a reader may too, and a tag or an action beat would fix it. In your own chapter, a mistake would sound like Mireille's voice saying one of Ines's lines.

What one voice misses and a cast catches

If you've had your chapter read aloud in one voice, you know what it catches: a repeated word, a missing "the", a sentence too long to say in one breath, dialogue that sounds stiff out loud. A full cast, with a voice for each character, catches those too. It also catches one thing a single voice can't: who is talking.

One voice says every line the same way, so a line readers would give to the wrong character still sounds fine. With a voice for each character, it sounds wrong the moment you hear it. That is the dialogue-tag test.

Listening won't catch everything. A voice says "their" and "there" the same way, and it can't hear formatting like italics, so still run spell check. Expressive voices also sometimes change a word, so still read the text once to check the exact words. While the sample above was being made, one take (one recording of a line) reworded the end of a narration sentence. Shortening the sentence fixed it.

The dialogue-tag test: is it clear who is speaking?

The rule: if the AI gives a line to the wrong character, read that line again, because a reader might make the same mistake. Before LyricWinter makes any audio, it splits your chapter into narration and dialogue and gives every line a speaker. A wrong speaker is a warning sign, not proof. The AI can miss something a careful reader would get.

Three places it usually trips, and the fix for each:

• A long back-and-forth with no tags. Partway through, add a tag or an action beat with the character's name, so a reader can find their place again.
• Two characters who are both "she" or both "he" in the same few lines. Use a name in the tag.
• A line of dialogue right after another character's action. Start a new paragraph, or tag the line.

How do I check who says each line for free?

Press Manual voice assignment, next to the Create audio button at the top of the section. On a phone, it's under that button's down arrow. It's free and makes no audio. Studio finds the characters and suggests a voice for each, then stops so you can check and change the voices yourself. To see every line, open the ⋯ menu (Studio tools) at the top right, choose Characters & voices, and switch to the Lines tab. Filter by a character to see only their lines.

When a line goes to the wrong character

First decide whether the page is unclear. Look at the sentence before the line. If it is unclear, fix your writing: add a tag, or give the speaker an action beat. That's the fix your readers need.

If the page is clear and the AI slipped, set the right speaker so the audio matches. Studio calls each line or stretch of narration a block. In the Lines tab, open the line's ⋯ menu, choose Block settings, pick the name under Speaker, and press Save speaker & block.

Bring in your chapter from wherever you write

Each chapter lives in its own Studio section. Paste the text, or open the ⋯ menu at the top right, choose Import story, and pick From link or From file.

Links work for chapters anyone can open without logging in. For an unposted chapter, or a work locked to logged-in AO3 users, paste the text.

• The AO3 import keeps only the story text. Your summary, notes and work header are left out, so they aren't read aloud. Turning posted AO3 chapters into MP3 files is covered in AO3 to MP3, and listening to fic on AO3 in AO3 text to speech.
• From file opens plain text (TXT), Markdown (.md, another plain text format), screenplay files (Fountain and Final Draft), EPUB ebooks, PDF, and photos of pages (JPG or PNG). In Google Docs on a computer, the Markdown download is under File, then Download.^[1] In Word, Save As or Save a Copy can make a PDF.^[2] On a phone, paste the text instead.
• On the free plan, one section can hold up to 30,000 characters of text (letters, spaces and punctuation), about 5,300 words. The paid plans hold more: 90,000 on MyWinter and 150,000 on UltraWinter. They also split longer pasted text into sections for you.

How do I hear my story read aloud with a voice per character?

LyricWinter Studio works in your web browser at lyricwinter.com/studio, with nothing to install. You sign in with Google or Apple. Making audio uses words, LyricWinter's credits for audio (see the cost section below).

1. Open a new project, name it, press New section, and paste your chapter. To bring in a file or a link instead, open the ⋯ menu at the top right and choose Import story.
2. Press Manual voice assignment. On a phone or other narrow screen, tap the small down arrow on the Create audio button and choose it there.
3. Check who says each line. Open the ⋯ menu at the top right, choose Characters & voices, and tap the Lines tab. Fix any line that went to the wrong character.
4. Check each character's voice in the Characters tab, next to Lines. Studio has already picked a voice for every speaker. If no voice fits a character, you can design a new AI voice for them at no cost in words. Each character also has an engine, and some engines use more words than others, as the cost section below shows.
5. Press Create audio. Before you press it, the button shows about how many words it will use. Auto emotion, in the down-arrow menu on that button, is on by default. It adds notes on how each line should sound, written the way each engine expects. To write your own, see the ElevenLabs v4 audio tags guide.
6. Listen with the playback bar, which appears at the bottom once there's audio, and read along as it plays. Its More audio actions menu (⋯) sets the speed, from 0.75× (slower) to 2× (twice as fast). Use normal speed (1×) the first time you listen, and try 1.25× once you know the chapter.
7. Edit as you go. When you're done, press the button again. It may now say Update audio instead of Create audio. It makes new audio only for the lines that need it, so you don't pay for the whole chapter again.
8. To redo one line, press its circular arrow in the Lines tab, then Regenerate in the New take panel. Older takes stay under Audio versions in the line's ⋯ menu, so you can keep the best one.

For more on how the cast is chosen, read how LyricWinter works.

How do I keep a character's voice the same in every chapter?

If Mireille sounds different in chapter 12 than she did in chapter 1, listeners notice. LyricWinter gives you two ways to lock a voice to a character.

A pin. Pin a voice to a character once, and it comes back whenever that name appears in a future project. The pin keeps the engine too. In Characters & voices, open the ⋯ menu next to the character's name, choose Voice settings and audio, then Pin as default. A pin matches the character's exact name, but capital letters don't matter. Pin Narrator too, to keep the same narrator voice. The voice pinning guide covers the details.

A Studio project. Put each chapter in the same project as its own section. The cast belongs to the project, so a voice or engine you set applies in every chapter. The Characters tab shows who is In this section and who appears Elsewhere in this project. A posted AO3 work can come in as a project in one step, and you choose which chapters to include before anything is created. To start one, open a new Studio project.

If no voice sounds like the character in your head, describe one. Studio's voice designer has a Voice direction box for the accent, age, tone and cadence (the rhythm of their speech). Designing a voice costs no words.

How do I fix how AI pronounces names I invented?

If the voices disagree about how to say a name, readers will too. AI voices guess at invented names, and the guess is useful: it's one way a reader might say the name. In the Marigold scene, the customs boat drifts toward Ceirnach Sound, a place that exists only in that story. It's meant to be said KEER-nakh. An early take started it with an S sound. The final take, with the same voice and engine, started it with a K.

Studio's Fix pronunciation tool lets you hear a name six ways from three engines and pick the one you meant:

1. Make the chapter's audio first, which uses words. Pronunciation samples need spoken audio in the section.
2. Select one word, such as "Ceirnach". Then open the ⋯ menu (Studio tools) at the top right and choose Fix pronunciation.
3. Check Word to fix. If the word appears more than once, choose Selected place or All matches.
4. Under How should it sound?, spell it the way you say it. The example in the box is "shuh-VAWN". Then press Prepare pronunciation.
5. Press Create pronunciation samples. Studio will "Compare six samples across Inworld, Cartesia, and Fish Audio," two from each engine. Each sample is charged only for that one word, so all six cost a handful of words.
6. Pick the sample you like. Then choose Change affected blocks (only the lines you're fixing) or Change speaker-wide (the whole character).

Applying a sample switches those lines, or the whole character, to the engine that made the sample.

Add a pronunciation note to your author's notes too, so readers who never press play say it your way.

How do I share my chapter with beta readers?

Beta readers read your draft before you post it. Yours don't need an account, an app or a file. Share a chapter from LyricWinter Studio, and they open a link, press play and read along. The page shows the full text with speaker labels and highlights the part being read. It also puts a play button beside every passage, so they can jump straight to the scene they have notes on.

1. Once the chapter has audio, open More audio actions (⋯) in the playback bar at the bottom, and choose Share listening link.
2. Press Create listening link, then Copy.

Edit the chapter, update the audio, and the same link plays the new version. The share window says it plainly: "One link stays current as you edit." You never resend a file. For a serial, use Share project from the project's menu instead: one link, with a Section picker for every chapter. A section's share window can also prepare an audio player to paste into your own website.

Anyone with the link can read and play the chapter. The page asks search engines like Google not to show it in their results. When the read is done, open the same share window and press Stop sharing, which turns the link off right away.

You own the audio you make

The audio you make is yours. You can post it or sell it, as long as you own or have permission to use the text and every voice in it. The full wording is in LyricWinter's Terms of Service. If you post your audio, a few habits help:

• Label it AI-generated. Say so in the title, summary or tags, so listeners know before they press play.
• On AO3, don't use the Podfic tag for AI audio. Podfic means fans recording stories in their own voices, so the tag is for human recordings. Describe your audio plainly in your other tags instead.
• Check your site's rules. As of October 2026, Royal Road's AI-text policy asks for an "AI-Generated" tag on stories written with AI and asks for "AI-Assisted" when AI edits heavily. It says nothing about audio.^[3] For SpaceBattles or any other site, read its pinned rules first.
• Check your contract. If a publisher owns the audio rights to your story, ask them before you post any audio version.
• Keep other people's fic private. If you make audio of someone else's story, read the author's notes for their wishes and keep the audio to yourself. AO3's FAQ says podfics of someone else's work can be posted only with permission from the copyright owner.^[4] For private listening, see how to listen to fanfiction.
• Selling an audiobook? Read the store's narration rules first. ACX, which publishes audiobooks to Audible, doesn't allow text to speech, AI or other automated narration without its approval.^[5]

How much does it cost to hear a chapter read aloud?

Finding who says each line is free, so the dialogue-tag test costs nothing. So are designing voices and AI sound effects. You pay only for spoken audio, in words (LyricWinter's credits). Words you buy never expire.

On most engines, a typical English chapter uses about one LyricWinter word for each word of story. The free plan includes 2,500 words a month, enough for about one 2,500-word scene on those engines. After that, every 2,500 words cost about $1, on the MyWinter plan ($6 a month for 15,000 words) or in word packs. So a 5,000-word chapter comes to about $2. ElevenLabs v4 is the exception after October 12, 2026: it uses three times as many words.

These are estimates as of October 1, 2026. The v4 change starts on October 12, 2026, at 00:00 UTC. UTC is world standard time, so in the US that is still October 11. When Studio casts a voice that supports ElevenLabs v4, it picks v4 for you. So check each character's engine in the Characters tab before you make audio for a long story.

A few things add to the total: redoing a line, pronunciation samples, and Auto emotion's directions on ElevenLabs v4 lines. The down arrow on the Create audio button shows what Auto emotion adds and lets you turn it off.

To hear a long manuscript many times, look at the UltraWinter plan ($20 a month for 72,500 words). On that plan, lines on the F5-TTS engine cost no words, for any voice that offers it. So you can hear the whole book again after every revision. F5-TTS isn't our strongest engine, but it's enough to hear repeated words and stiff lines. See pricing for current plans.

Is LyricWinter right for you?

If your chapters have more than one speaker, yes. You get a voice for every character, the same voices chapter after chapter, invented names said the way you meant, and a link your beta readers can open without an account. Three things to know before your first chapter:

• LyricWinter Studio is in beta: it's new, and its screens and labels may still change.
• Links import from AO3, FanFiction.net, SpaceBattles and Substack. For Royal Road, a draft, or a Word file, paste the text.
• Export final mix, in the playback bar's menu, saves the finished audio one section at a time. You get an MP3 or WAV audio file, or SRT captions (a subtitle file).

Start with the free part. Open a project, add a chapter, press Manual voice assignment, and see who says each line before you spend any words. If a line goes to the wrong character, you found it before your readers did.

Run the free dialogue-tag test

Writing a serial or a whole book? Bring it in as one project, one section per chapter, as the import table above shows. For a shorter overview of checking a draft by ear, see manuscript audio proofing. Writing a script? See the screenplay table read page.

Text to speech for writers: FAQ

Read the full Terms of Service.