FACE TO FACE
Understand the person across from you.
Spoken input and translated output for in-person exchanges on one device.
HOME / PRODUCTS / VOXLINGO
VOXLINGO / MULTILINGUAL COMMUNICATION
Speak yours.
Be heard in theirs.
Speech recognition, translation and speech synthesis run as one chain on the device, for face-to-face conversations, meetings and calls.
01 / PRODUCT EXPERIENCES
Make room for everyone in the conversation.
MEETINGS
Keep the discussion moving.
Recognition and translation for multilingual meetings, as text or speech in your application.
CALLS
Bring translation into the call.
Speech translation across your call audio path, turn-taking and output.
These are integration scenarios. Available workflows and supported language pairs are confirmed during device and product evaluation; the illustration is not a live demonstration.
02 / POWERED BY VOXEDGE
One chain, from speech to speech.
01
Capture
Prepare microphone or application audio for speech processing.
02
Recognize
Turn spoken input into text with on-device ASR.
03
Translate
Translate text with the device-side multilingual model.
04
Speak
Synthesize translated text for spoken output.
TRANSLATION
103
Broad text coverage
On-device translation languages and writing variants.
RECOGNITION
36
Speech input
On-device ASR languages across the platform.
SYNTHESIS
45
Speech output
On-device TTS languages across the platform.
Translation, ASR and TTS coverage are counted separately. End-to-end speech translation requires support across the selected input language, translation direction and output voice.
03 / DEPLOYMENT
Choose the footprint for your device.
COMPACT MODEL
239 MiB
For tighter device budgets
239 MiB package; approximately 320 MiB peak memory. Measured decoding speed of 121 tokens per second on Helio G85.
BALANCED MODEL
548 MiB
For higher translation quality
548 MiB package. Internal evaluation reports +17.5% relative BLEU improvement versus Hy-MT2-1.8B (1.25-bit).
INTEGRATION
Evaluate the full experience
Align languages, audio routing, output mode, memory and power requirements before choosing the model and capability chain.
Balanced-model comparison: all 32 shared English-to-target directions, internal offline evaluation. BLEU measures translation quality; +17.5% is a relative improvement, not accuracy. Compact-model performance is configuration-specific and does not measure end-to-end speech translation latency.
04 / QUESTIONS
Plan your language experience.
Can translation run without a network?
The translation model runs on the device. A complete offline speech experience also needs the required ASR and TTS models installed and supported in your configuration.
Are all 103 languages supported for voice translation?
No. The 103 figure covers text translation languages and writing variants. Speech recognition and speech synthesis have separate coverage; the supported voice translation pairs must be checked across all three capabilities.
Can we integrate this into a calling or meeting app?
These are target integration scenarios. We evaluate your application’s audio access, languages, output behavior and device resources before confirming the supported workflow.