Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/phpaisdk/elevenlabs/llms.txt

Use this file to discover all available pages before exploring further.

The ElevenLabs provider integrates with the portable Generate::speech() API, letting you swap TTS providers without rewriting application code. ElevenLabs-specific controls — voice settings, normalization, pronunciation dictionaries, and output format strings — are passed through providerOptions('elevenlabs', [...]) so the portable interface stays clean while full provider fidelity is preserved.

​
Installation

composer require aisdk/elevenlabs
Set your API key as an environment variable, or configure the provider directly at boot time: 
use AiSdk\ElevenLabs;

ElevenLabs::create([
    'apiKey' => 'xi-...',
]);

​
Basic Usage

1

Build the speech request

Call Generate::speech() with the text you want to synthesize, then chain .model(), .voice(), .format(), and any provider options before calling .run().
use AiSdk\ElevenLabs;
use AiSdk\Generate;

$result = Generate::speech('Welcome to the show.')
    ->model(ElevenLabs::model('eleven_flash_v2_5'))
    ->voice('JBFqnCBsd6RMkjVDRZzb')
    ->format('mp3')
    ->run();
2

Save the output

The result exposes an AudioData object. Call save() with a filesystem path to write the audio file.
$result->output->save(__DIR__ . '/speech.mp3');

​
Voice IDs

ElevenLabs requires a voice ID for every TTS request. Pass it with the portable .voice() method. If you omit the voice, the adapter throws an InvalidArgumentException before the request is sent. 
Generate::speech('Good evening.')
    ->model(ElevenLabs::model('eleven_v3'))
    ->voice('JBFqnCBsd6RMkjVDRZzb')   // ElevenLabs voice ID — required
    ->run();
You can also supply voice_id inside providerOptions('elevenlabs', [...]), but the portable .voice() method takes precedence.

​
Output Formats

Use .format() to select an audio format with a portable key. The adapter maps it to the appropriate ElevenLabs format string automatically.
.format() valueElevenLabs format stringDescription
mp3 (default)mp3_44100_128MP3 at 44.1 kHz, 128 kbps
wavwav_44100WAV at 44.1 kHz
pcmpcm_44100Raw PCM at 44.1 kHz
ulawulaw_8000µ-law at 8 kHz (telephony)
alawalaw_8000A-law at 8 kHz (telephony)
Generate::speech('Hello.')
    ->model(ElevenLabs::model('eleven_flash_v2_5'))
    ->voice('JBFqnCBsd6RMkjVDRZzb')
    ->format('wav')
    ->run();
When you need an ElevenLabs-specific format string that has no portable equivalent — such as mp3_44100_192 or pcm_22050 — set output_format in providerOptions('elevenlabs', [...]). It overrides the value derived from .format().
->providerOptions('elevenlabs', ['output_format' => 'mp3_44100_192'])

​
Provider Options

providerOptions('elevenlabs', [...]) accepts the full set of documented ElevenLabs request fields. The adapter routes them to the correct part of the HTTP request automatically.

​
Query Parameters

The following fields are sent as URL query parameters, not in the request body, matching the ElevenLabs API contract.
​
output_format
string
An ElevenLabs-specific output format string such as mp3_44100_192 or pcm_22050. Overrides the value derived from the portable .format() method.
​
enable_logging
bool
When false, ElevenLabs does not store the request in your history. Sent as enable_logging=false in the query string.
​
optimize_streaming_latency
int
Deprecated ElevenLabs latency-optimization level (0–4). Sent as a query parameter. Prefer eleven_flash_v2_5 for low latency instead.

​
JSON Body Fields

All other fields are merged into the JSON request body alongside text and model_id.
​
voice_settings
object
Fine-grained voice control. All sub-fields are optional and override the voice’s saved defaults.
  • stability (float) — Consistency of delivery. Higher values are more predictable; lower values are more expressive.
  • similarity_boost (float) — How closely the output resembles the original voice.
  • style (float) — Stylistic intensity. Only effective on v2 and later models.
  • use_speaker_boost (bool) — Applies a speaker-similarity enhancement pass.
​
apply_text_normalization
string
Controls how ElevenLabs normalizes text before synthesis. Accepted values are auto, on, and off.
​
language_code
string
A BCP-47 language code hint (e.g. en, fr, de). Improves pronunciation for multilingual models such as eleven_multilingual_v2.
​
seed
int
Integer seed for deterministic generation. Using the same seed and input should produce the same audio output.
​
pronunciation_dictionary_locators
array
Array of pronunciation dictionary locators to apply. Each entry should be an object with pronunciation_dictionary_id and optionally version_id.

​
Voice Settings Example

use AiSdk\ElevenLabs;
use AiSdk\Generate;

$result = Generate::speech('Welcome to the show.')
    ->model(ElevenLabs::model('eleven_flash_v2_5'))
    ->voice('JBFqnCBsd6RMkjVDRZzb')
    ->format('mp3')
    ->providerOptions('elevenlabs', [
        'voice_settings' => [
            'stability'        => 0.5,
            'similarity_boost' => 0.75,
            'style'            => 0.2,
            'use_speaker_boost' => true,
        ],
        'apply_text_normalization' => 'auto',
        'language_code' => 'en',
        'seed' => 42,
    ])
    ->run();

$result->output->save(__DIR__ . '/speech.mp3');

​
Full Example with All Options

use AiSdk\ElevenLabs;
use AiSdk\Generate;

$result = Generate::speech('Hello there.')
    ->model(ElevenLabs::model('eleven_flash_v2_5'))
    ->voice('JBFqnCBsd6RMkjVDRZzb')
    ->providerOptions('elevenlabs', [
        // Query parameters
        'output_format'              => 'wav_44100',
        'enable_logging'             => false,
        'optimize_streaming_latency' => 2,

        // JSON body fields
        'seed' => 42,
        'voice_settings' => [
            'stability'        => 0.5,
            'similarity_boost' => 0.75,
        ],
    ])
    ->run();

$result->output->save(__DIR__ . '/speech.wav');

​
Available Model IDs

eleven_flash_v2_5

Ultra-low-latency model. Best for real-time applications and streaming scenarios.

eleven_v3

Latest high-quality model. Best for expressive, studio-quality voice output.

eleven_multilingual_v2

Multilingual model supporting 29 languages. Use with language_code for best results.
For the lowest possible latency, use eleven_flash_v2_5. For the highest quality output — podcast narration, audiobooks, or voice-over work — use eleven_v3.

Build docs developers (and LLMs) love

Get started for freeTalk to us