The Rufa model family.
Seven models, one API: Rufa hears speech, speaks text in natural voices, translates between 100+ languages, understands what was said and holds live voice conversations.
Rufa Realtime (V1 · Beta)
Speech to speech. Talk to AI the way you talk to a person.
Rufa Realtime listens, thinks and answers out loud in one live conversation. You speak, it starts answering in a fraction of a second, in your language, and you can interrupt it at any time. Choose Realtime Flash for instant replies, or Realtime Brain (Rufa Brain 31B) for deeper answers and long instructions.
- Input
- Live speech (browser or phone line)
- Output
- Live speech and a transcript
- Starts answering
- Around 200 ms (Flash) · under 500 ms (Brain)
- Best for
- Call assistants, reception, voice agents
- Realtime Flash and Realtime Brain
- Interrupt at any time
- Follows you when you switch language
Rufa Ears
Speech to text. Speech recognition for Lëtzebuergesch, the way it is really spoken.
Rufa Ears turns Luxembourgish speech into clean, timed text: phone calls, meetings, interviews and voice notes. It is tuned on the language as people actually speak it, including the French, German and English words that come with it.
- Input
- Audio and phone calls
- Output
- Text with timed segments
- Languages
- Luxembourgish
- Best for
- Calls, meetings, subtitles
- Timed segments for subtitles
- Long recordings, split automatically
- Live and recorded audio
Rufa Ears Global
Speech to text. One model for the many languages of Europe.
Rufa Ears Global listens in Luxembourgish and the main European languages, so multilingual teams and customers are understood without switching tools.
- Input
- Audio
- Output
- Text with timed segments
- Languages
- Luxembourgish, English, French, German, Portuguese, Spanish, Italian, Dutch
- Best for
- Multilingual call centres and media
- Multilingual speech
- Same API as Rufa Ears
- Timed segments
Rufa Speak
Text to speech. Natural voices that speak their own language.
Rufa Speak reads any text in a natural, expressive voice. Each voice in the library belongs to a language and speaks it like a native. Pick a speaking role, such as reading, news, conversation, phone call or phone menu, and Rufa Speak adapts its delivery.
- Input
- Text, up to 5,000 characters
- Output
- Natural speech (MP3 / WAV)
- Voices
- 1000+ voices in 28 languages
- Best for
- Voice agents, phone menus, audio content
- Speaking roles: read, news, chat, call, phone menu
- Speed control
- Voice cloning with recorded consent (coming soon)
Rufa Voice Clone
Text to speech. Your own voice, ready to speak any text.
Rufa Voice Clone learns a voice from a short recording and speaks any text in it with Rufa Speak. It only clones a voice with the speaker’s recorded consent, and the voice stays yours: private to your account, removed whenever you ask.
- Input
- About 30 seconds of clear speech
- Output
- A private voice for Rufa Speak
- Consent
- Recorded from the speaker, always
- Best for
- Your brand’s voice, narration, accessibility
- Ready in about a minute
- Noise cleaned from the recording
- Private to your account
Rufa Translate
Language model. Translate Luxembourgish, and 100+ more languages.
Rufa Translate turns text or speech in one language into another, from Lëtzebuergesch to English, French, German, Portuguese and 95+ more. It detects the language for you, keeps names, numbers and formatting, and a Rufa voice of the target language reads the result out.
- Input
- Text or speech
- Output
- Translated text, natural speech (MP3)
- Languages
- 100+ to translate; voices in 28
- Best for
- Customer messages, documents, live help
- Language detection
- Speak instead of typing
- Hear and download the translation
Rufa Flash
Language model. The fast thinker for live conversations.
Rufa Flash understands Luxembourgish and answers within a fraction of a second. It powers live conversations and processes calls as they happen: translation, replies, intents and next steps.
- Input
- Text and transcripts
- Output
- Answers, translations, structured data
- Speed
- Built for real-time
- Best for
- Voice agents and live call assistance
- Luxembourgish to English translation
- Live replies for voice agents
- Structured JSON output
Rufa Brain
Language model. Deeper reasoning over every conversation.
Rufa Brain reads whole calls and documents and returns what matters: summaries, corrections to transcripts, layouts by speaker and the answers your team needs. Rufa Flash is built for speed; Rufa Brain is built for depth.
- Input
- Transcripts and long text
- Output
- Summaries, corrections, insights
- Strength
- Context and accuracy
- Best for
- Call reviews, reports, quality checks
- Transcript correction
- Speaker layout and summaries
- Company-aware answers