Batch Text to Speech: Generate Many Voice Clips at Once

Batch text to speech lets you turn a list of scripts into many separate audio clips in one go, using the same voice and settings for each. It saves hours when you have lesson libraries, product descriptions, game dialogue or notification sounds to produce. Voxmint supports batch generation of up to 100 lines per request in the developer API, and the same consistent-voice approach works across everything you generate in the Studio.

What is batch text to speech?

Instead of pasting one script, generating, downloading and repeating, you send a list of lines with one set of settings. Each line becomes its own clip. If one line fails, only that line reports a problem and the others still run, so a typo in line 47 does not cost you the other 99.

When batch generation is the right tool

E-learning libraries: one clip per lesson step, per slide or per quiz question.
Product and catalogue audio: short spoken descriptions for many items.
App prompts and notifications: greetings, confirmations and error messages that must all sound like the same speaker.
Game and interactive dialogue: many short lines from one character.
Language courses: vocabulary lists with one clip per word or phrase.
Localisation: the same set of lines in several of the 30 supported languages.

If your project is one long narration, generate it section by section in the Studio instead. Batch is for many independent short clips.

Why a consistent voice matters

Each catalogue voice is fixed to the same speaker, so line 1 and line 100 sound like the same person even when generated weeks apart. This is the main reason to use a catalogue voice for a batch: the voice becomes a stable part of your product rather than something that shifts between takes.

If you need a voice that is not in the catalogue, use voice design to describe one in words, covering age, accent, mood and pace, and save the description to your library to reuse.

Step by step: plan a batch

1. Collect and clean your lines

Put each line on its own row in a spreadsheet or a text file. Remove duplicates, fix typos and decide how numbers, dates and abbreviations should be read. Cleaning the text now is much cheaper than regenerating later.

2. Keep lines short and self-contained

Short lines generate faster and are easier to fix. Each line should make sense by itself, because each becomes a separate clip. If a sentence depends on the one before it, keep them in the same line.

3. Pick one voice and one set of settings

Choose the voice, language and any style prompt, such as slower or warmer. Settings apply to the whole batch, so test them on a few representative lines first.

4. Run a pilot of five to ten lines

Pick a mix: a short line, a long line, one with numbers, one with a name, one with a question. Listen to all of them. Fix pronunciation by adjusting spelling, and settle on the style before running everything.

5. Run the full batch

Submit your lines. In the API, a batch accepts up to 100 lines; split longer lists into several requests and keep a record of which lines went into which.

6. Review the results line by line

Check the status of each line. Listen to a sample, then spot-check the rest. Re-run only the lines that failed or need changes.

7. Name and organise the output

Use a naming scheme that ties each clip back to its source line, such as a line number or an ID from your spreadsheet. Future you will be grateful.

Using batch in your own app

The developer API accepts API keys, and a batch request takes a list of lines plus one shared set of settings: the mode (speech or design), the voice and so on. The result reports each line separately. A sensible workflow:

Build the list of lines from your database.
Send them in groups of up to 100.
Store each returned clip with its source ID.
Retry only the lines that failed.

For live features where speed to first sound matters, such as reading a message aloud as it arrives, use streaming generation instead of batch. For captions or read-along, word timings can be requested for a generation as well. Batch is best for work you can prepare ahead of time.

Tips for consistent quality across a batch

Same punctuation style everywhere. If some lines end with full stops and others do not, pacing will vary.
Standardise numbers and abbreviations. Choose one way to write them and apply it to every line.
Keep a pronunciation list. When you fix a name's spelling for one line, apply the same fix to every line where it appears.
Test the longest and shortest lines. Extremes show problems first.
Don't change settings mid-project. If you must, regenerate the earlier lines too so the set stays uniform.

Common mistakes

Running 100 lines before a pilot. Always test a small sample.
Mixing very different line types. A greeting, a legal notice and a joke call for different delivery; consider separate batches with different style prompts.
Forgetting the text limit. Very long scripts should be split into shorter lines.
Losing the source link. Without IDs, matching clips to lines becomes guesswork.

Batch across languages

With 30 languages in the catalogue, you can run the same list of lines through several languages, using a voice native to each. Translate carefully first, since a literal translation often sounds stiff. Then pilot each language separately, because pacing and punctuation conventions differ. Language pages such as Japanese text to speech or Spanish text to speech show the voices available.

FAQ

How many lines can I send in one batch?

Up to 100 lines per request in the developer API. For longer lists, send several requests.

What happens if one line fails?

The failure is reported against that line only, and the other lines still run. Re-run just the failed ones.

Will every clip sound like the same speaker?

With a catalogue voice, yes. The voice is fixed to the same speaker, so clips stay consistent across lines and across days.

Can I use different voices in the same batch?

A batch applies one set of settings to all lines. To use several voices, send one batch per voice.

Have a long list of lines to voice? Try batch generation with Voxmint and keep every clip in the same voice.