Long Text to Speech - 50,000 Characters at Once

Long text to speech: paste up to 50,000 characters, get context-aware chunked generation merged into one audio file with a full subtitle track.

最終更新: 2026-10-18

New: Audio Drama Studio — paste a story and get a full-cast audio drama, cast automatically with every line directed; the same cast can perform it in English, Spanish, Japanese and more. Hear finished pieces in the Theater.

Long text to speech means handing a whole chapter, lecture script or podcast episode to an AI voice in one go: up to 50,000 characters per job, returned as one MP3 with one subtitle track for the whole text. Regular text to speech stops at 5,000 characters a request, which leaves you splitting, generating and stitching by hand; long-form mode does the splitting, the transitions and the merge for you.

The hard part of long text to speech isn't "long"

Anyone can split 10,000 words into pieces and generate each one. The hard part is making the seams inaudible: intonation must not reset, and the tone at the end of one chunk has to carry into the next. Long-form mode sends each chunk to the model together with its neighbors as context, so transitions stay continuous.

Real sample: three paragraphs, one file

A 197-character fiction excerpt, split into 3 paragraphs by blank lines, narrated by a Mandarin broadcast voice and merged into one 49.7-second file:

中国語50秒

原稿:深夜十一点，林夏终于合上了最后一页实验记录。窗外的雨已经停了，城市像是屏住了呼吸。她知道明天的发布会将决定三年的心血是归零还是改变一切。电话在这时响起，屏幕上是一个她以为永远不会再出现的名字。她撑着，她看了怔怔奇妙，然后接了起来。对方没有汗旋，只说了一句话：“他们提前了。”林夏方下电话，从抽屉深处取出那枚芯片，直见炳两，心跳却出奇地平稳。她对着空荡荡的实验室轻声说：“这一次，轮到我先出牌了。

Here is the opening of a full chapter: narration handing over to a character and back, sentence after sentence, the tone carried all the way to the last line. What long-form mode has to prove is not that it can read, but that a few hundred words in it is still the same voice:

ボイス: George英語34秒

原稿:Chapter One. Winter came early that year. By the time the first snow fell, the old ferryman had not taken his boat out for three days. Some said the river had frozen; some said he was ill. Only he knew that he was waiting for someone. On the eighth day of the last month, the snow stopped. A young man in a black cloak appeared at the crossing, said nothing, and set a single copper coin on the bow. The ferryman glanced at it, untied the rope, and as the boat left the bank he said quietly: your father crossed this river the same way.

Want a whole chapter rather than an excerpt? Here is Chapter Two in full — one submission, one generation, narration and dialogue trading off for the entire length:

ボイス: George英語133秒

原稿:Chapter Two. The Crossing. On the ninth day of the last month, snow began again at first light. The old ferryman tied the boat to its post and went inside to build a fire. He did not ask the young man's name, and the young man did not offer it. They sat across a low table while the charcoal cracked in the brazier. "The year your father crossed," the ferryman said, "the snow was like this." The young man looked up. "You knew him?" "I did. He sat where you are sitting, put a copper coin on the table, and said he was going over the mountain to find someone." The young man said nothing. He drew a letter from inside his coat; the envelope had worn pale at the edges, but the seal was unbroken. "He never came back," he said. "He left this. I have never opened it." The ferryman glanced at the letter, then at the snow beyond the window. "The road over the mountain cannot be walked in heavy snow. You will have to wait." "How long?" "Until you have decided whether to open it." The room fell quiet. The coals dimmed; the young man added two pieces and the fire came back. He remembered, suddenly, his father crouched at the stove in the same way, feeding it wood and telling stories about the far side of the mountain: a bridge, and on the bridge a man who waited for another man who never came. "He told me about a bridge," the young man said. The ferryman nodded. "The bridge is still there. So is the one who waits." When the light had fully come, the snow stopped. The young man stood, tucked the letter back into his coat, and pushed open the door. Thin ice floated on the river; the far mountain was white enough to hurt the eyes. "I won't open it," he said. "I'll ask him." The ferryman did not stop him. He only handed over the pole. "Take the boat yourself. On the far bank, tie it to the third post. That is the one your father used." As the boat left the shore the ferryman stood at the crossing, as he had thirty years before, and watched until it became a black dot and vanished into the glare of the snow.

The same chapter in Chinese, submitted in one go:

ボイス: 老者说书人中国語45秒

原稿:第一章。那年冬天来得早,第一场雪落下的时候,渡口的老船夫已经三天没有开船了。有人说河面结了冰,有人说他病了,只有他自己知道,他在等一个人。腊月初八,雪停了。一个披着黑斗篷的年轻人出现在渡口,什么也没说,只把一枚铜钱放在船头。老船夫看了那铜钱一眼,解开缆绳。船离岸的时候,他轻轻说了一句:你父亲当年,也是这样过河的。

深夜十一点，林夏终于合上了最后一页实验记录。……
(11 p.m. Lin Xia finally closed the last page of the lab notes…)

电话在这时响起。……
(That's when the phone rang…)

林夏放下电话，从抽屉深处取出那枚芯片。……
(She hung up and took the chip from the back of the drawer…)

Listen for the transition from paragraph one to two — that's a chunk boundary, and the delivery doesn't reset.

What long text to speech does

Auto-chunking: splits on blank lines, then on sentence ends for oversized paragraphs, keeping every chunk under 2,500 characters
Context continuity: each chunk is generated with the previous and next text, so intonation carries over
Merged output: all chunks concatenated into one MP3; character timestamps shifted by cumulative duration into one subtitle track spanning the whole text
One charge: credits computed per character and shown before you generate; if any chunk fails, the whole job refunds

Splitting a book longer than 50,000 characters

A full book usually runs past the per-job limit, so submit it in parts:

One chapter per job. Chapters are the natural break: one MP3 and one subtitle file each, and fixes happen chapter by chapter.
Combine short chapters. If chapters are only a few thousand characters, put several into one job, as long as the total stays under 50,000.
Leave blank lines between paragraphs. Chunking follows blank lines, so clean paragraphs mean chunk boundaries fall between paragraphs rather than mid-sentence.
Keep one voice for the whole book. Pick the same voice for every job so forty chapters sound like one narrator.

Long text to speech or another tool?

Your content	Use	Limit per request
Ads, social clips, short scripts	Text to speech	5,000 characters
A chapter, lecture or episode in one voice	Long text to speech	50,000 characters
Two or three people trading lines	Multi-speaker dialogue	2,000 characters
A story with a different voice per character	Audio drama studio	Cast automatically

What it's for

Audiobooks & web novels: submit whole chapters, narrated by a library anchor voice or your cloned narrator
Lecture scripts: one file per lesson with full subtitles; edit one paragraph, regenerate just that part
Podcast scripts: write, generate, publish — no booth
Multilingual long content: generate each translated version with a native voice for that language

Cost

Long-form jobs are billed per character: 70 credits per 1,000 characters on the standard model, 40 on the fast model. The page shows the credits before you generate, and a failed job is refunded automatically. To try it once, the $3.90 trial pack has 150 credits that never expire. See pricing.

How to use it

Voice Studio → Long-form: paste up to 50,000 characters; the page shows how many chunks and how many credits in real time. Pick a voice and model (expand advanced settings for stability and speed), then generate. Roughly 1–2 minutes per 10,000 characters. Download the audio and the full SRT from the history panel when done.

FAQ

How much text can long text to speech take at once? Up to 50,000 characters per job. The text is split on blank lines, then at sentence ends for oversized paragraphs, into chunks of under 2,500 characters, and merged back into one MP3 with one subtitle track. For a book longer than that, submit it a few chapters at a time.

How is it different from regular text to speech? Regular text to speech takes up to 5,000 characters per request, which suits short scripts. Long text to speech takes up to 50,000, chunks the text for you, generates each chunk with its neighbours as context so the delivery does not reset, and returns one file with one subtitle track.

What does a long-form job cost? It is billed per character: 70 credits per 1,000 characters on the standard model, 40 on the fast model. The page shows the chunk count and the credits before you generate, and if any chunk fails the whole job is refunded.

How long does it take? Roughly 1–2 minutes per 10,000 characters. Download the audio and the full SRT from the history panel when it finishes.

Paste your longest chapter and try long text to speech on it:

Generate my first chapter →