Which languages can it translate between?

English, French, German, Spanish and Portuguese, in any direction: French to German, Spanish to English, and so on. That makes twenty language pairs. You don't need to say which language the speaker is using. It is detected automatically, so they can just start talking.

Can I get the translation as text only?

Yes. You can get the translation as text, or have it spoken out loud in the other language. When it is spoken, you still get the text too, so you can show captions while the voice plays.

Can I choose the voice the translation speaks in?

Yes, you pick one for every session. It can be any voice from our catalogue in the target language, or a clone of your own voice. With a clone, the speaker is heard in another language and still sounds like themselves.

About 3 seconds on average from someone speaking to hearing the translation, across all language pairs. OpenAI GPT Realtime takes 3.6 seconds. Most systems work in three steps, one after the other: write down what was said, translate the text, then speak it. Ours translates while it listens, so there is one step fewer to wait for.

We tested it on everyday conversation, like people talking about work, travel or the weather, rather than on scripted text. For every translation we check two things: whether the words are right, and whether the meaning is right. It does better than Google Gemini 3.5 on both, and better than OpenAI GPT Realtime on the words while matching it on the meaning. https://gradium.ai/blog/launching-gradium-translate.

Does it work on phone calls?

Yes. It accepts the audio formats phone systems use, so it plugs into an existing phone line without converting the audio first. It also knows when the speaker has finished talking, so the translation comes at the right moment in the conversation.

Spotlight post

Gradium Extends Funding to $100 Million and Expands to Silicon Valley

Gradium extends its seed funding to $100 million, welcoming new investors including NVIDIA, and opens a San Francisco Bay Area office to scale its real-time voice AI.

July 8, 20263 min read

Gradium Extends Funding to $100 Million and Expands to Silicon Valley

Models

Text-to-SpeechTurn text into natural sounding voice Speech-to-TextTranscribe audio into text instantly Live TranslationConvert voice to voice in real time Voice CloningInstance or pro voice clone Voice DesignA brand-new voice from a text prompt On-Device TTSOffline real-time natural voice

Showcase

Agent DemoTest our voice agent live GradbotBuild voice agents in a single prompt CommunityBuilders, researchers and creators App GalleryApps built on Gradium

Hear them in your language,while they are still speaking.

Hold a conversation in a language you don’t speak.

Translate to

Start translating Read the docs

No credit card required

The numbers behind live translation

20 pairs
Every direction covered
Any of five languages into any other, in any direction.
400+ voices
Any voice, or your own
Every output voice in the catalogue, or a clone of your own.
3.0 s
End-to-end latency
Average end-to-end latency across all language pairs.

Try now Talk to us

No credit card required

Fewer models, faster translation

The standard stack transcribes, then translates the text, then speaks it. Ours translates as it listens so one whole model and its handoff leave the critical path.

Launching Gradium Translate

Models in the path

Average end to end

3.0 s

Intermediate transcript

None

No marketing spin,just code and hard numbers

Paste one of these prompts into your coding agent and it will read the guide, install the SDK and open the stream. Nothing else to do, you're ready to go.

1Add Gradium live speech-to-speech translation to this project.
2
3Read https://docs.gradium.ai/guides/speech-to-speech-overview first and
4follow it over anything you assume. Use the official Gradium SDK for this
5codebase's language. Read the API key from an environment variable, never
6hardcode it.
7
8Then implement the flow, in this order:
9
101. Open one s2s_realtime session with model_name "s2s-translate",
11   stt_model_name "stt-translate", tts_model_name "default", the target
12   language in json_config, and a voice_id that exists in that target
13   language. voice_id is required, not optional.
142. Stream source audio in as PCM frames in one task and call send_eos when it
15   ends.
163. Read the socket in a second task. Play audio messages as they arrive and
17   collect text messages as a running translated transcript. Stop on
18   end_of_stream.
194. Make the target language and the voice ID configurable per session, not
20   per build, so a call can be re-pointed without a redeploy.
21
22Keep it small: one module, typed, with clear errors when the voice ID does
23not match the target language, when the socket drops, or when the key is
24missing. Do not add a UI unless I ask for one.

Speech in, speech out

No distortion under load

Latency stays flat as concurrency climbs, flexible deployment that fits your setup.

Real-time API

Bidirectional WebSocket streaming. Start sending before ready returns; first audio in ~200 ms. Telephony formats built in.

in · pcm 24 kHz

CLIENT/AGENT

WEBSOCKET

Gradium API

wss://api.gradium.ai/
api/speech/s2s

BidirectionalTelephony ulaw / alaw~200 ms

out · pcm 48 kHz

VOICE ENGINE

Copied!

Predictable & secure

Flat latency under load, enterprise SLAs, zero data retention, ISO 27001 certified.

1 conversationQuiet hour

I’d like to move my appointment.

50ms

Of course. Which day works?

10 000 conversationsPeak

I’d like to move my appointment.

50ms

Of course. Which day works?

Runs where you need it

Cloud API, dedicated instances, self-hosted, on-prem for data sovereignty, and cloud-provider marketplaces.