Models & Languages Overview

Models

Model	Description & Use
Flux TTS	Our latest-generation, conversation-native voice model — streaming-first, turn-based, and expressive by default. Recommended for all new builds: voice agents, customer service, IVR, and general-purpose synthesis. Served on `/v2/speak`.
Aura-2	Our previous-generation model, with voices across seven languages. Served on `/v1/speak`.
Aura	Our first-generation text-to-speech model. English voices only. Served on `/v1/speak`.

Start with Flux TTS. It’s our best-sounding, most accurate model and the default recommendation for every use case it serves.

The language comes from the voice’s model string — the language code at the end selects it (flux-desislava-bg is Bulgarian, flux-haley-en is English). A model is required on every request — Flux TTS on /v2/speak, Aura-2 and Aura on /v1/speak.

Example

To request a voice, set the model to the voice you want. Aura-2 and Aura use /v1/speak:

cURL (Aura-2, /v1/speak)

curl "https://api.deepgram.com/v1/speak?model=aura-2-thalia-en" \
  -H "Authorization: Token YOUR_DEEPGRAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "text": "Hello, how are you?" }' \
  --output audio.mp3

Flux TTS uses /v2/speak (streaming WebSocket and batch REST). See Getting Started with Flux TTS.

Replace YOUR_DEEPGRAM_API_KEY with your Deepgram API Key.

Flux TTS

Flux TTS is Deepgram’s best text-to-speech model and the recommended choice for every use case — real-time voice agents, customer service, IVR, and pre-rendered audio alike. It’s conversation-native: streaming-first, turn-based, expressive by default, and consistent across turns, with native interruption handling and strong accuracy on the strings that trip up agents in production (alphanumerics, drug names, and other hard-to-say entities). Model strings follow the format flux-{voice}-{language} (e.g. flux-alexis-en), and model is required on every /v2/speak connection.

Model Option	Language
`flux-{voice}-en`	English (American, British, Irish, Australian, Indian, Singaporean, Filipino): `en`
`flux-{voice}-bn`	Bengali: `bn`
`flux-{voice}-bg`	Bulgarian: `bg`
`flux-{voice}-ca`	Catalan: `ca`
`flux-{voice}-zh`	Chinese (Mandarin, Taiwanese Mandarin, Cantonese): `zh`
`flux-{voice}-hr`	Croatian: `hr`
`flux-{voice}-cs`	Czech: `cs`
`flux-{voice}-da`	Danish: `da`
`flux-{voice}-nl`	Dutch (Dutch, Flemish): `nl`
`flux-{voice}-et`	Estonian: `et`
`flux-{voice}-fi`	Finnish: `fi`
`flux-{voice}-fr`	French: `fr`
`flux-{voice}-de`	German (German, Swiss German): `de`
`flux-{voice}-el`	Greek: `el`
`flux-{voice}-gu`	Gujarati: `gu`
`flux-{voice}-hi`	Hindi: `hi`
`flux-{voice}-hu`	Hungarian: `hu`
`flux-{voice}-id`	Indonesian: `id`
`flux-{voice}-it`	Italian: `it`
`flux-{voice}-ja`	Japanese: `ja`
`flux-{voice}-kn`	Kannada: `kn`
`flux-{voice}-ko`	Korean: `ko`
`flux-{voice}-lv`	Latvian: `lv`
`flux-{voice}-lt`	Lithuanian: `lt`
`flux-{voice}-ms`	Malay: `ms`
`flux-{voice}-mr`	Marathi: `mr`
`flux-{voice}-no`	Norwegian: `no`
`flux-{voice}-pl`	Polish: `pl`
`flux-{voice}-pt`	Portuguese (Brazil, Portugal): `pt`
`flux-{voice}-ro`	Romanian: `ro`
`flux-{voice}-ru`	Russian: `ru`
`flux-{voice}-sk`	Slovak: `sk`
`flux-{voice}-sl`	Slovenian: `sl`
`flux-{voice}-es`	Spanish: `es`
`flux-{voice}-sv`	Swedish: `sv`
`flux-{voice}-tl`	Tagalog: `tl`
`flux-{voice}-ta`	Tamil: `ta`
`flux-{voice}-te`	Telugu: `te`
`flux-{voice}-tr`	Turkish: `tr`
`flux-{voice}-uk`	Ukrainian: `uk`
`flux-{voice}-ur`	Urdu: `ur`
`flux-{voice}-vi`	Vietnamese: `vi`

Text normalization (numbers, dates and abbreviations read out) is available for English, German, Dutch, French, Hindi, Italian, Japanese, Portuguese, Russian and Spanish; the other 32 languages are in Early Access without it — write numbers and dates out in words.

See the full Flux TTS Voices & Languages catalog for every voice and its characteristics.

Aura-2

Aura-2 is Deepgram’s previous-generation text-to-speech model, with voices across seven languages. For new work, build on Flux TTS. Model strings follow the format aura-2-{voice}-{language} (e.g. aura-2-thalia-en).

Model Option	Language
`aura-2-{voice}-en`	English (American, British, Australian, Filipino): `en`
`aura-2-{voice}-es`	Spanish (Mexican, Peninsular, Colombian, Argentine, Latin American): `es`
`aura-2-{voice}-de`	German: `de`
`aura-2-{voice}-fr`	French: `fr`
`aura-2-{voice}-nl`	Dutch: `nl`
`aura-2-{voice}-it`	Italian: `it`
`aura-2-{voice}-ja`	Japanese: `ja`

See the full Aura Voices & Languages catalog for every voice, accent, and audio sample. Select Spanish voices (Aquila, Carina, Diana, Javier, Selena) support English–Spanish codeswitching.

Aura

Aura is Deepgram’s first-generation text-to-speech model. English voices only, served on /v1/speak.

Model Option	Language
`aura-{voice}-en`	English (American, British, Irish): `en`

See the Aura Voices & Languages catalog for the full list.