Introduction
Send up to 10,000 characters to elevenlabs/v4-turbo/text-to-speech for app prompts, narration drafts or localized voice-over. Choose the language, normalization mode and output format, then review the completed audio before publication.
| Field | Value |
|---|---|
| Input | Text |
| Output | Audio |
| Speech API identifier | elevenlabs/v4-turbo/text-to-speech |
| Text ceiling | 10,000 characters per request |
| Casting | 21 premade voices; default Roger |
| Format setting | 9 choices; default |
| Recorded test conditions | November 13, 2026: 75 characters, Sarah, |
Key features & highlights
A short Sarah recording, with context
Our 75-character Turbo request returned a 3.40-second MP3 after about 7.6 seconds on November 13, 2026. The call used Sarah and mp3_44100_128; it provides a concrete integration reference without predicting other scripts or workloads.
Cast a prompt set from 21 voices
Roger is the default among 21 premade voices, with Sarah, Daniel and Lily as other examples. Pair a selected voice with auto language detection or one of 91 language codes, then audition the actual wording of each localized prompt.
Put a delivery cue beside the words
Square-bracket tags such as [laughs] or [whispers] belong inside text. Our test used one tag, so judge the returned performance rather than assuming that a larger collection of directions has been validated.
Choose from 9 output options
Output defaults to mp3_44100_128, and mp3_22050_32 is another available choice. Select a listed format that your player accepts, then keep that setting fixed while you compare script revisions.
Request parameters
| Parameter | Required | Description |
|---|---|---|
text | Yes | Provide the required spoken script, up to 10,000 characters. Put inline cues such as [whispers] at the relevant words and proofread the text before generating a larger batch. at least 1 characters · at most 10000 characters |
voice_id | No | Choose one of the 21 premade voices; omitting the selection uses Roger. Audition Sarah, Daniel or Lily with the same short line rather than supplying a custom voice ID. |
language | No | Select auto or one of the 91 supported language codes. For a short localized prompt, choose its intended language and provide text already written in that language. |
stability | No | Use stability to tune consistency of delivery, then listen to a short sample. Keep the script and voice unchanged while comparing settings; no value guarantees matching takes. |
similarity_boost | No | Adjust |
output_format | No | Select from 9 options, with |
text_normalization_mode | No | Choose auto, on or off to control normalization of numbers and abbreviations. |
clone_strength | No | Leave |
seed | No | Keep a fixed seed when you want to compare one script or setting change at a time. Treat the result as a new performance rather than expecting identical audio. |
Pricing
$0.048 / 1,000 characters
Charges follow the characters in text, including spaces, punctuation and inline tags. Voice, language and output format leave the per-character rate unchanged; failed requests are not charged. Check the price displayed on this page.
Use cases
App status messages
Input "Your upload is ready. Open the preview to continue." with Sarah to obtain a spoken status cue for an app. Audition it in the actual screen flow before reusing the voice.
Product-demo drafts
Submit "Choose a workspace, add your files, then open the preview." to generate a narration draft. Compare the returned audio with the demo timing before preparing a final take.
Course instructions
Use "Pause the lesson and list three examples from your own project." as a short course-audio script. The result is a single-voice recording you can review alongside the exercise.
Localized release prompts
Supply "Tu archivo está listo." with language es to create a Spanish app voice prompt. Prepare each translated script yourself and generate a separate recording for it.
Game character lines
Write "[whispers] The gate is open. Follow the lanterns." to try a directed NPC line. Receive one voice performance, then listen for the cue and the clarity of the instruction.
Prompt guide & pro tips
Draft one Turbo line
Put a short, proofread script in text and choose a single premade voice for it.
Submit the model and input
Use elevenlabs/v4-turbo/text-to-speech as model in the createTask request, with your speech fields inside input.
Listen to the completed asset
Poll recordInfo with the returned taskId, download the audio asset after success and check its pronunciation in your player.
ElevenLabs v4 Turbo API for short speech scripts
ElevenLabs v4 Turbo API renders text as a single premade-voice recording with ElevenLabs' Eleven v4 Turbo speech model. Use it for app notices, narration drafts and multilingual voice-over, with up to 10,000 characters in each request.
ElevenLabs says Turbo targets interactive applications in its launch announcement. The endpoint described here uses a submitted task and a completed audio asset; our measured task time below is not a streaming latency claim.
Submit a speech request
Create a task with model set to elevenlabs/v4-turbo/text-to-speech and the speech fields inside input. Send POST https://api.spicyapi.ai/api/v1/jobs/createTask with Authorization: Bearer YOUR_SPICY_API_KEY and Content-Type: application/json.
{
"model": "elevenlabs/v4-turbo/text-to-speech",
"input": {
"text": "[whispers] Your preview is ready. Open it when you have a moment.",
"voice_id": "Sarah",
"language": "en",
"output_format": "mp3_44100_128"
}
}
This request illustrates the field layout rather than reproducing our measured script. Roger is the default voice when you omit voice_id; a custom voice ID is not accepted here.
Collect the response
Poll GET /api/v1/jobs/recordInfo?taskId=YOUR_TASK_ID with the task ID returned at submission. Keep the ID to check the same job.
A shortened submission response can look like this:
{
"code": 200,
"data": {
"taskId": "YOUR_TASK_ID",
"state": "queued"
}
}
A shortened successful task response exposes the audio asset:
{
"code": 200,
"data": {
"taskId": "YOUR_TASK_ID",
"state": "succeeded",
"output": {
"assets": [
{"url": "https://example.com/turbo-result.mp3", "mime": "audio/mpeg"}
]
}
}
}
These excerpts use placeholder IDs and URLs. Download from data.output.assets after success and audition the file in your intended player.
Parameters and endpoint scope
The input controls cover text, premade voice selection, language, delivery, normalization, output format and seed. Use the listed choices for each control rather than importing fields from another speech API.
| Parameter | Supported input or purpose | Practical choice |
|---|---|---|
text | Required; up to 10,000 characters | Proofread the words and inline cues |
voice_id | 21 premade voices; default Roger | Try Sarah, Daniel or Lily on a short line |
language | auto or one of 91 language codes | Match the language of your prepared script |
stability | Delivery consistency control | Compare a change with the voice held fixed |
similarity_boost | Chosen-voice similarity control | Listen to the same phrase after adjustment |
clone_strength | No effect on premade voices | Omit for this endpoint's voice set |
text_normalization_mode | auto, on or off | Check number and abbreviation readings |
output_format | 9 choices; default mp3_44100_128 | mp3_22050_32 is another MP3 option |
seed | Supports controlled comparisons | Do not expect identical recordings |
Character-based pricing
Text characters determine billing for this speech endpoint, rather than seconds of generated audio. Use the current price shown on this page to estimate a batch from its scripts; do not infer a charge from the duration of the returned MP3.
Improve delivery before scaling up
Short audition scripts let you hear pronunciation and performance before recording a larger collection. For an app notice, try “Your upload is ready. Open the preview to continue.” and listen to the transition between its two sentences.
- Start with one voice and the untagged script to establish a reference.
- Add a cue such as
[whispers]or[laughs]where the delivery changes, then listen again. - Compare normalization settings on actual numbers or abbreviations from your project.
- Keep the voice, format and seed fixed while changing one element of the script or delivery controls.
A fixed seed makes comparisons easier without promising identical output. Prepare translated wording before selecting its language code; language selection is not a translation step.
What our calls measured
Our November 13, 2026 Turbo call returned a 3.40-second MP3 in about 7.6 seconds from 75 characters with Sarah. The test used one inline tag; a wider tag vocabulary was not validated.
| Call | Text characters | Voice | Output format | Audio duration | Time to finish |
|---|---|---|---|---|---|
| v4 Turbo | 75 | Sarah | mp3_44100_128 | 3.40 seconds | About 7.6 seconds |
| v4 | 84 | Daniel | mp3_22050_32 | 5.17 seconds | About 4.7 seconds |
These calls are integration observations, not a controlled comparison: the scripts, voices and output formats differ. The v4 call also specified English and seed 7; neither result predicts performance for your next script.
Limits, errors and model selection
Keep text within 10,000 characters and use a listed premade voice when checking a rejected request. This endpoint does not yet expose multi-speaker dialogue input or custom voice IDs, and clone_strength cannot change its premade voices.
If a call fails, check the returned error and your submitted fields before trying a revised request. If the audio completes but a number sounds wrong, test a short extract with different normalization settings and explicitly written wording.
Choose Turbo for drafts, high-volume prompt libraries and workflows that prioritize latency. Compare ElevenLabs v4 for final narration takes using the same script and settings; consider a different speech endpoint when custom voices or dialogue input are requirements, after checking that endpoint's documented support. Before publishing a Turbo performance, confirm your rights to its text and voice and avoid impersonating any real person.
Publish scripts and use voices for which you hold the necessary rights. Do not use a generated performance to impersonate a real person.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "elevenlabs/v4-turbo/text-to-speech",
"input": {
"text": "Welcome back. Your files finished rendering a minute ago and are ready to download.",
"language": "auto",
"voice_id": "Roger",
"output_format": "mp3_44100_128",
"text_normalization_mode": "auto"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.




