Deepgram

Authentication

Deepgram requires an API key, which you can create from the Deepgram console. Configure it in your environment file:

1

DEEPGRAM_API_KEY=your-deepgram-api-key

The key is sent as Authorization: Token <key> on every request.

Alternatively, you can omit the API key and pass an HTTP client that is already configured with a base URI and the Authorization header. In that case the endpoint argument is ignored, as the pre-configured client is used as-is.

Text-to-speech

Pass the voice model name and a Text instance:

1
2
3
4
5
6
7

use Symfony\AI\Platform\Bridge\Deepgram\Factory;
use Symfony\AI\Platform\Message\Content\Text;

$platform = Factory::createPlatform(apiKey: $_ENV['DEEPGRAM_API_KEY']);

$result = $platform->invoke('aura-2-thalia-en', new Text('Hello world'));
file_put_contents('/tmp/out.mp3', $result->asBinary());

Audio knobs supported by Deepgram (encoding, container, sample_rate, bit_rate) are forwarded as query-string parameters when you pass them as invocation options:

1
2
3
4
5

$platform->invoke(
    'aura-2-thalia-en',
    new Text('Hello world'),
    ['encoding' => 'linear16', 'sample_rate' => 24000],
);

1
2
3
4
5
6
7
8
9

use Symfony\AI\Platform\Result\Stream\Delta\BinaryDelta;

$result = $platform->invoke('aura-2-thalia-en', new Text('Hello world'), ['stream' => true]);

foreach ($result->asStream() as $chunk) {
    if ($chunk instanceof BinaryDelta) {
        echo $chunk->getData();
    }
}

Speech-to-text

Pass an Audio instance. The bridge uploads the audio as raw bytes with the right Content-Type header — files are streamed from disk, so even large recordings do not have to fit in memory:

1
2
3
4
5
6
7

use Symfony\AI\Platform\Bridge\Deepgram\Factory;
use Symfony\AI\Platform\Message\Content\Audio;

$platform = Factory::createPlatform(apiKey: $_ENV['DEEPGRAM_API_KEY']);

$result = $platform->invoke('nova-3', Audio::fromFile('/path/to/audio.mp3'));
echo $result->asText().PHP_EOL;

Deepgram STT options (language, smart_format, punctuate, diarize, utterances…) are forwarded as query parameters:

1
2
3
4
5

$platform->invoke(
    'nova-3',
    Audio::fromFile('/path/to/audio.mp3'),
    ['smart_format' => 'true', 'language' => 'en', 'diarize' => 'true'],
);

Use 'language' => 'multi' to enable multilingual (code-switching) transcription on nova-3 models.

1
2
3
4
5
6
7

$platform->invoke(
    'nova-3-general',
    [
        'type' => 'input_audio',
        'input_audio' => ['url' => 'https://example.com/audio.mp3'],
    ],
);