Voices at 90% below ElevenLabs.

Natural, expressive voices that stream as they render. Voice cloning from a 10-second clip. Works with the ElevenLabs and OpenAI SDKs.

pip install elevenlabs

from elevenlabs import ElevenLabs

client = ElevenLabs(
    api_key=os.environ["MSERVE_API_KEY"],

− ~~base_url="https://api.elevenlabs.io",~~+

)

stream = client.text_to_speech.convert_as_stream(
    voice_id=voice_id, text=reply,
)

The right model for each voice.

Preset voices run on Kokoro in eight languages. Cloned voices run on Chatterbox. Voice conversion turns any recording into either.

Upload a clip, ten seconds is enough, and get a persistent voice you can use forever. Clone only voices you have permission to use.

Audio for the first sentence starts while the rest renders. WebSockets and streaming HTTP, in the formats you already use.

Model	Best for	License	Price
`Kokoro 82M`	Preset voices	Apache-2.0	$3.00/ 1M chars
`Chatterbox`	Voice cloning	MIT	$6.00/ 1M chars
`Chatterbox Voice Conversion`	Voice conversion	MIT	$0.06/ minute
`Whisper Large v3 Turbo`	Transcription	MIT	$0.006/ minute

Which formats do you return?

mp3, opus, pcm, wav, ulaw, and alaw.

How is Voice billed?

Per character for speech, per minute for conversion and transcription. No seats. Purchased credits never expire.

Can I bring my ElevenLabs voice IDs?

Premade ElevenLabs voice ids work unchanged. Custom voices: clone them here from a short clip and you get a new persistent voice ID.

One email when your key is ready. Nothing else.