Skip to content

Voices and models

Every text-to-speech call needs a voice ID, and most accept a model ID. You can hard-code both, but listing them at runtime lets users pick a voice and keeps you from sending a model a setting it doesn't support.

Listing voices

voices.list() returns one page at a time. The default page holds 10 voices, and you can ask for up to 100:

val page = elevenLabs.voices.list(
    ListVoicesRequest {
        pageSize = 25
        search = "narrator"
    },
)

page.voices.forEach { voice ->
    println("${voice.name} (${voice.id}) ${voice.labels}")
}

search matches free text. voiceType and category are passed to ElevenLabs as-is, so use the values from the ElevenLabs voices API.

To fetch the next page, pass the token from the current one back in:

suspend fun allVoices(): List<Voice> {
    val voices = mutableListOf<Voice>()
    var request = ListVoicesRequest { pageSize = 100 }
    while (true) {
        val page = elevenLabs.voices.list(request)
        voices += page.voices
        val token = page.nextPageToken
        if (!page.hasMore || token == null) return voices
        request = request.toBuilder().apply { nextPageToken = token }.build()
    }
}

totalCount is null unless you set includeTotalCount = true. Leave it off if you only page forward.

Each Voice has a previewUrl when ElevenLabs provides one. It points at a short sample you can play in a voice picker without spending characters on generation.

Getting one voice

val voice = elevenLabs.voices.get("JBFqnCBsd6RMkjVDRZzb")

JBFqnCBsd6RMkjVDRZzb is one of the ElevenLabs premade voices, which is why the examples in these docs use it. Any voice ID from voices.list() works the same way.

Listing models

models.list() returns every model your account can use, in a single call:

val ttsModels = elevenLabs.models.list().filter { it.canDoTextToSpeech }

The capability flags tell you what a model accepts before you send a request:

Property Meaning
canDoTextToSpeech The model can generate speech.
canUseStyle The model accepts VoiceSettings.style.
canUseSpeakerBoost The model accepts VoiceSettings.useSpeakerBoost.
maximumTextLengthPerRequest The longest input one request may contain, when ElevenLabs reports it.
languages The languages the model supports.

The model list changes rarely, so fetching it once per app launch is usually enough.

Model.id is a ModelId, so you can pass it straight to a request's modelId. ModelId also has constants for the models this SDK version knows about, such as ModelId.ElevenFlashV2_5. For a newer model, construct one from its ID: ModelId("eleven_new_model").