README

Official ElevenLabs generative-media provider for the PHP AI SDK.

Installation

composer require aisdk/elevenlabs

Configuration

Set ELEVENLABS_API_KEY, or configure the provider directly:

use AiSdk\ElevenLabs;

ElevenLabs::create([
    'apiKey' => 'xi-...',
    'baseUrl' => 'https://api.elevenlabs.io/v1',
    'headers' => ['X-Custom-Header' => 'value'],
]);

ELEVENLABS_BASE_URL overrides the default base URL when no baseUrl option is supplied.

Supported SDK Capabilities

Capability	Support
Speech generation	Native
Transcription	Native
Realtime transcription	Native through `Live::transcribe()`
Text, image, embeddings, video	Not provided by this package

Model IDs are opaque. Use the ElevenLabs model identifier appropriate for the operation, such as eleven_flash_v2_5, eleven_v3, or scribe_v2.

Provider-owned extensions cover the direct ElevenLabs creative-media APIs that do not belong in core:

Extension	Surface
Voice changer and isolation	Audio-to-audio transforms
Voice design	Design previews, remix a voice, and save a selected generated preview
Music	Compose, composition plans, detailed output, video-to-music, upload, and stem separation
Sound effects and dialogue	Text-to-audio generation, including dialogue timestamps
Dubbing	Create a dub, inspect its status, and retrieve dubbed media or transcripts
Forced alignment	Align supplied audio and transcript text

These extensions are available from either the configured provider instance or the static facade:

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

$music = $elevenLabs->music()->compose('A warm acoustic intro.');
$sameResource = ElevenLabs::music();

Speech Generation

Voice IDs are an ElevenLabs requirement, so provide one using the portable voice() method.

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,
        ],
    ])
    ->run();

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

Use providerOptions('elevenlabs', ...) for documented ElevenLabs request fields such as apply_text_normalization, language_code, seed, and pronunciation dictionaries. Set output_format there when an ElevenLabs-specific output format is needed.

The adapter sends output_format, enable_logging, and the deprecated optimize_streaming_latency field as query parameters, while keeping voice settings and generation controls in the JSON request body, matching the ElevenLabs API contract.

Transcription

use AiSdk\Content;
use AiSdk\ElevenLabs;
use AiSdk\Generate;

$result = Generate::transcription(Content::audio(__DIR__.'/meeting.mp3'))
    ->model(ElevenLabs::model('scribe_v2'))
    ->providerOptions('elevenlabs', [
        'language_code' => 'en',
        'diarize' => true,
    ])
    ->run();

echo $result->output->text;

Realtime Transcription

Scribe v2 Realtime uses the core Live API. Install aisdk/transport for the ready-made WebSocket transport:

composer require aisdk/transport

use AiSdk\ElevenLabs;
use AiSdk\Live;
use AiSdk\Live\TranscriptCompleted;
use AiSdk\Live\TranscriptUpdate;
use AiSdk\Transport;

$session = Live::transcribe()
    ->model(ElevenLabs::model('scribe_v2_realtime'))
    ->language('en')
    ->audioFormat('pcm_16000')
    ->providerOptions('elevenlabs', [
        'commit_strategy' => 'manual',
        'include_timestamps' => true,
    ])
    ->connect(Transport::auto());

$session->sendAudio($pcmBytes);
$session->commitAudio();

foreach ($session->events() as $event) {
    if ($event instanceof TranscriptUpdate) {
        echo "\r{$event->text}"; // Revisable partial transcript.
    }

    if ($event instanceof TranscriptCompleted) {
        echo "\n{$event->text}\n";
    }
}

aisdk/transport is optional. Without it, pass any application transport implementing AiSdk\Live\Contracts\TransportInterface; provider event encoding and normalization still come from this package.

For a browser connection, create a short-lived single-use token on your server and return only its value to your authenticated client:

$secret = Live::transcribe()
    ->model(ElevenLabs::model('scribe_v2_realtime'))
    ->clientSecret();

return ['token' => $secret->value, 'expires_at' => $secret->expiresAt];

The browser connects natively to ElevenLabs using that token. Never expose the workspace API key to client-side code.

const { token } = await fetch('/api/elevenlabs/scribe-token').then(response => response.json());
const query = new URLSearchParams({
    model_id: 'scribe_v2_realtime',
    audio_format: 'pcm_16000',
    commit_strategy: 'vad',
    token,
});
const socket = new WebSocket(`wss://api.elevenlabs.io/v1/speech-to-text/realtime?${query}`);

socket.addEventListener('message', ({ data }) => {
    const event = JSON.parse(data);

    if (event.message_type === 'partial_transcript') {
        console.log('Partial:', event.text);
    }

    if (event.message_type === 'committed_transcript') {
        console.log('Final:', event.text);
    }
});

When include_timestamps=true, the ordinary committed transcript is normalized as TranscriptCompleted; the following committed_transcript_with_timestamps message remains available as a raw ProviderEvent, preserving ElevenLabs word metadata without duplicating the portable completion event.

ElevenLabs Media Services

These capabilities are intentionally provider-owned instead of being added to aisdk/core.

Voice Changer

$result = ElevenLabs::voiceChanger()->convert(
    voiceId: 'JBFqnCBsd6RMkjVDRZzb',
    audio: Content::audio(__DIR__.'/performance.wav'),
    options: [
        'model_id' => 'eleven_multilingual_sts_v2',
        'remove_background_noise' => true,
        'output_format' => 'wav_44100',
    ],
);

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

Voice Isolator

$result = ElevenLabs::voiceIsolator()->isolate(
    Content::audio(__DIR__.'/noisy-interview.mp3'),
);

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

Music and Sound Effects

$music = ElevenLabs::music()->compose('Warm acoustic guitar intro with gentle rain.', [
    'model_id' => 'music_v1',
    'music_length_ms' => 30_000,
]);

$effect = ElevenLabs::soundEffects()->generate('A wooden door slowly creaking open.');

Music also supports composition plans and detailed metadata without opting into an HTTP streaming variant:

$plan = ElevenLabs::music()->plan('A tense cinematic cue with a quiet ending.', [
    'model_id' => 'music_v2',
    'music_length_ms' => 30_000,
]);

$track = ElevenLabs::music()->composeFromPlan($plan, [
    'model_id' => 'music_v2',
    'output_format' => 'mp3_48000_192',
]);

$detailed = ElevenLabs::music()->composeDetailed('A bright synth-pop theme.', [
    'model_id' => 'music_v2',
    'with_timestamps' => true,
]);

echo $detailed->songId;
$detailed->output->save(__DIR__.'/theme.mp3');

Direct music transforms use typed Content inputs:

$soundtrack = ElevenLabs::music()->videoToMusic([
    Content::video(__DIR__.'/opening.mp4'),
    Content::video(__DIR__.'/closing.mp4'),
], [
    'description' => 'Cinematic, restrained, then uplifting.',
    'tags' => ['cinematic', 'orchestral'],
    'model_id' => 'music_v2',
]);

$stems = ElevenLabs::music()->separateStems(
    Content::audio(__DIR__.'/song.mp3'),
    ['stem_variation_id' => 'six_stems_v1'],
);

$stems->save(__DIR__.'/stems.zip');

music()->upload() uploads source audio for ElevenLabs composition-plan and inpainting workflows and returns its typed song ID, optional plan, and optional word timestamps.

Text to Dialogue

$dialogue = ElevenLabs::dialogue()->create([
    ['text' => '[giggling] Knock knock.', 'voice_id' => 'JBFqnCBsd6RMkjVDRZzb'],
    ['text' => '[curious] Who is there?', 'voice_id' => 'Aw4FAjKCGjjNkVhN1Xmq'],
], ['model_id' => 'eleven_v3']);

Use withTimestamps() when character alignment and per-voice segments are needed:

$dialogue = ElevenLabs::dialogue()->withTimestamps([
    ['text' => 'Ready?', 'voice_id' => 'JBFqnCBsd6RMkjVDRZzb'],
    ['text' => 'Ready.', 'voice_id' => 'Aw4FAjKCGjjNkVhN1Xmq'],
]);

$dialogue->output->save(__DIR__.'/dialogue.mp3');
foreach ($dialogue->voiceSegments as $segment) {
    echo "{$segment->voiceId}: {$segment->start} - {$segment->end}\n";
}

Voice Design and Remixing

Voice design is a two-step generation flow: generate previews, then save the selected preview as a usable voice.

$design = ElevenLabs::voiceDesign()->design(
    'A warm, expressive documentary narrator with measured pacing.',
    [
        'model_id' => 'eleven_ttv_v3',
        'auto_generate_text' => true,
    ],
);

$preview = $design->previews[0];
$preview->audio->save(__DIR__.'/voice-preview.mp3');

$voice = ElevenLabs::voiceDesign()->create(
    generatedVoiceId: $preview->generatedVoiceId,
    name: 'Documentary narrator',
    description: 'A warm, expressive documentary narrator with measured pacing.',
);

echo $voice->id;

To transform an eligible existing voice instead, call voiceDesign()->remix($voiceId, $description, $options) and save one of its generated previews in the same way.

Dubbing

Create a dub from local audio/video or from a URL-backed Content value, inspect the asynchronous job, then retrieve the output:

$job = ElevenLabs::dubbing()->create(
    Content::video(__DIR__.'/interview.mp4'),
    targetLanguage: 'es',
    options: ['source_lang' => 'en'],
);

$status = ElevenLabs::dubbing()->status($job->id);

if ($status->status === 'dubbed') {
    ElevenLabs::dubbing()
        ->output($job->id, 'es')
        ->save(__DIR__.'/interview-es.mp4');

    $subtitles = ElevenLabs::dubbing()->transcript($job->id, 'es', 'srt');
    file_put_contents(__DIR__.'/interview-es.srt', $subtitles->content ?? '');
}

Forced Alignment

$alignment = ElevenLabs::forcedAlignment()->create(
    Content::audio(__DIR__.'/narration.wav'),
    'The exact transcript spoken in the recording.',
);

foreach ($alignment->words as $word) {
    echo "{$word->text}: {$word->start} - {$word->end}\n";
}

Scope

This package deliberately stays on the direct generative/media surface.

Included	Deliberately excluded
TTS, batch STT, and realtime Scribe	ElevenAgents and its telephony/runtime management APIs
Voice changer, isolation, design, and remixing	Voice-library browsing, cloning/training, and general voice CRUD
Music generation and direct music transforms	Music finetune, asset/history, and marketplace management
Sound effects and text-to-dialogue	Studio, Audio Native, Flows, and project/editor management
Dubbing creation and result retrieval	Dubbing Studio resource editing, listing, and deletion
Forced alignment	Account, workspace, API-key, billing, analytics, and administrative APIs

Saving a generated voice preview is included because it is the required second step of the Voice Design and Remix APIs. Pronunciation-dictionary management, AI-audio detection, human production services, and HTTP streaming variants of otherwise one-shot media endpoints are outside this package surface. Core realtime transcription remains fully supported through Live::transcribe().

Testing

composer test:all

The default suite is fixture- and conformance-based. Credentialed Live network verification is separate and is not run by composer test.

Documentation

Community

Contributing
Support
For private security reports, email security@phpaisdk.com.

aisdk / elevenlabs

Maintainers

Package info

Fund package maintenance!

Statistics

Security