Laravel AI SDK Integration

VOICEVOX for Laravel integrates with Laravel AI SDK, so you can generate speech through the Audio facade. Two drivers are available:

voicevox-client (client mode)
voicevox (native mode)

Client driver (`voicevox-client`)

Sends HTTP requests to the official VOICEVOX engine. No FFI required.

Prerequisites

docker pull voicevox/voicevox_engine:cpu-latest
docker run --rm -p '127.0.0.1:50021:50021' voicevox/voicevox_engine:cpu-latest

Configuration

Add voicevox-client provider to config/ai.php.

'providers' => [
    'voicevox-client' => [
        'driver' => 'voicevox-client',
        'key' => env('VOICEVOX_URL', 'http://127.0.0.1:50021'),
    ],
],

Usage

use Laravel\Ai\Audio;

$response = Audio::of('I love coding with Laravel.')
    ->voice('ずんだもん')
    ->generate('voicevox-client');

Storage::put('talk.wav', $response->content());

Native driver (`voicevox`)

Calls VOICEVOX CORE directly via FFI. Only models loaded in core.vvms in config/voicevox.php can be used.

Configuration

'providers' => [
    'voicevox' => [
        'driver' => 'voicevox',
    ],
],

Usage

use Laravel\Ai\Audio;

$response = Audio::of('ネイティブで話すのだ')
    ->voice('ずんだもん')
    ->generate('voicevox');

Storage::put('talk.wav', $response->content());

Values accepted by `voice()`

voice() accepts:

VOICEVOX style ID (numeric string)
Character alias (for example, ずんだもん, 四国めたん/ノーマル)

When omitted, default-female (ID: 10) is used.

Alias table

Alias	Style ID	Character
`ずんだもん`	1	Zundamon (Sweet)
`ずんだもん/あまあま`	1	Zundamon (Sweet)
`ずんだもん/ノーマル`	3	Zundamon (Normal)
`ずんだもん/セクシー`	5	Zundamon (Sexy)
`ずんだもん/ツンツン`	7	Zundamon (Tsundere)
`ずんだもん/ささやき`	22	Zundamon (Whisper)
`ずんだもん/ヒソヒソ`	38	Zundamon (Murmur)
`四国めたん/あまあま`	0	Shikoku Metan (Sweet)
`四国めたん`	2	Shikoku Metan (Normal)
`四国めたん/ノーマル`	2	Shikoku Metan (Normal)
`四国めたん/セクシー`	4	Shikoku Metan (Sexy)
`四国めたん/ツンツン`	6	Shikoku Metan (Tsundere)
`四国めたん/ヒソヒソ`	37	Shikoku Metan (Murmur)
`春日部つむぎ`	8	Kasukabe Tsumugi (Normal)
`波音リツ`	9	Naminori Ritsu (Normal)
`雨晴はう`	10	Amehare Hau (Normal)
`玄野武宏`	11	Kurono Takehiro (Normal)
`白上虎太郎`	12	Shirakami Kotaro (Normal)
`青山龍星`	13	Aoyama Ryusei (Normal)
`冥鳴ひまり`	14	Meinei Himari (Normal)
`九州そら`	16	Kyushu Sora (Normal)
`default-female`	10	Amehare Hau (Normal)
`default-male`	12	Shirakami Kotaro (Normal)

use Laravel\Ai\Audio;

Audio::of('Test')->voice('3')->generate('voicevox-client');

You can also inspect available speakers via Voicevox::speakers().

use Revolution\Voicevox\Voicevox;

$speakers = Voicevox::speakers();

Agent (KanalizerAgent)

Native mode does not provide enable_katakana_english, so you can use KanalizerAgent to convert English text into katakana first.

use Revolution\Voicevox\Ai\Agents\KanalizerAgent;
use function Revolution\Voicevox\talk;

$kana = KanalizerAgent::make()->prompt('Convert English to katakana with KanalizerAgent');

$response = talk($kana['kana'] ?? $kana->text, id: 1)->generate(id: 1);

$response->storeAs('native', 'kanalizer.wav');

AI-based katakana conversion is not always perfect. Review generated text before synthesis in production workflows.

Client: Talk

Review the standard client-mode API flow.

Native: Talk

Review direct FFI-based synthesis flow.