Skip to content

Intelligence,
on your device.
No Cloud Bill. Faster Than A Blink. Barely Any Battery.

On-device voice AI with no cloud bill, speech latency of ~70 ms and barely any battery use.

NO CLOUD BILL

Runs without a network.

FASTER THAN A BLINK

On-device speech latency.

BARELY ANY BATTERY

Extra battery use / 12 hours.

[ 01 / PRODUCTS ] → VOXEDGE

The engine behind every voice.

Listening, translating and speaking, all on the device: VAD, ASR, MT and TTS, built to run within the power budget of real devices.

Get started ↗

VoxEdge

01 / NO CLOUD BILL
$0per callcloud cost

Recognition, translation and speech run on the device.

0 network round trips
02 / FASTER THAN A BLINK
~70mson-device speech latency

Speech turns into text as fast as people talk, with no waiting on a network.

8.02% English word error rate¹
8.91% Chinese character error rate¹
03 / BARELY ANY BATTERY
180mAh / 12hestimated incremental use²

Low-power capture and local transcription, optimized across hardware and software.

12 h estimated consumption²
04 / MULTILINGUAL REACH
103languages & writing variants³

Translation coverage, with speech focused on underserved local languages.

36 On-device ASR languages
45 On-device TTS languages
COMPATIBILITY MediaTek Helio G81 and above Compatible down to entry-level chips, so more platforms can run it.
[ + ] TEST NOTES · INTERNAL TESTS ON MTK9600 · CONFIGURATION-SPECIFIC
  1. Internal noisy-speech evaluation at ≥10 dB signal-to-noise ratio; lower is better. Proprietary always-on ASR, 65M parameters; VoxSense reference package approximately 80 MiB.
  2. Estimated incremental battery use: 180 mAh over 12 hours, based on 15 mAh per hour. Reference setup: continuous speech playback; transcription once per minute.
  3. 103 refers to on-device translation languages and writing variants. ASR and TTS coverage are listed separately; availability varies by configuration.

The diagrams above are illustrative: a typical blink is about 200 ms; the battery fill shows about 96.4% remaining after this feature’s estimated extra use alone: 15 mAh × 12 h = 180 mAh, or 3.6% of an example 5,000 mAh battery. Other phone battery use is excluded.

TRANSLATION QUALITY · +17.5% relative BLEU vs. Hy-MT2-1.8B (1.25-bit) on all 32 shared English-to-target directions; 20 of 21 mainstream directions won vs. Google Translate offline. BLEU relative improvement, not accuracy. Internal evaluation.

Learn more ↗
[ 01 / PRODUCTS ] → VOXSENSE

From sound. To intelligence.

The promise you made. The detail you forgot. Captured the moment it’s said, always on, at low power.

Get started ↗

VoxSense

A voice-activity detector wakes the pipeline only when someone speaks. On-device speech recognition turns it into local text. What your product does with that text is up to you.

01 / SENSE → 02 / TRANSCRIBE → 03 / UNDERSTAND → 04 / ASSIST
Low-power captureOn-device transcriptionContext & memory
180 mAh / 12h Estimated 12-hour incremental battery use, based on 15 mAh per hour. MTK9600; continuous speech, transcription once per minute.
Learn more ↗
VOXSENSE / CONTEXT ILLUSTRATIVE UI
09:41LOCAL ●
Keep what matters.
Illustrative context & memory experience
JUST NOW / A COMMITMENT“Remind me to reply to that email when I get to the office.”
RELATED CONTEXTConversation → Transcript → Memory
LOW-POWER CAPTURE → LOCAL TEXTCONTEXT → ASSISTANCE
[ 01 / PRODUCTS ] → VOXLINGO

Speak yours. Be heard in theirs.

Face to face or on a call, speech is translated and spoken back on the device, no network needed.

Get started ↗

VoxLingo

Hear. Translate. Speak.

Speech recognition (ASR), machine translation (MT) and speech synthesis (TTS) run as one chain on the device, from spoken input to translated speech, for face-to-face talk and calls.

Speech & text translationLocal inferenceMultilingual speech

Made for meetings and everyday conversations, on phones, earbuds and smart glasses.

Learn more ↗
People with diverse skin tones meeting around a table with a VoxLingo smartphone at its center.
[ 03 / PARTNERS ] → ECOSYSTEM

Built with partners.
From silicon to cloud.

Chipset vendors, device brands and cloud providers, working with us to bring on-device voice and language AI into products people use every day.

[ 04 / LANGUAGES ] → LOCAL LANGUAGE. GLOBAL REACH.

More local languages.
More people included.

Speech and translation for markets across Africa and South Asia, in the languages and accents people actually speak.

  • Kiswahili
  • हिन्दी
  • Hausa
  • বাংলা
  • Bahasa Indonesia
  • தமிழ்
  • English
  • Chinese
  • and more
[ 05 / INTEGRATION ] → FROM APPLICATION TO SDK

Four steps.
Simple to integrate.

// 001

Create an application

Enter your app’s platform and package name, then choose the capabilities and languages to try.

// 002

Approval and NDA

Wait for approval and sign the NDA.

// 003

Download the SDK

Build and download the SDK for your platform. It includes every capability in your license.

// 004

Initialize with your key

Pass your application key at start-up. The device activates against the license on first use.

BUILD WITH VOXNEXUS

Your next product.
[ on-device AI ]

NO CLOUD BILL · FASTER THAN A BLINK · BARELY ANY BATTERY

CONTACT US ↗