NATHAN AI · MICRO / MACRO · LOCAL FIRST · PRESERVE THE ORIGINAL

THE DEVICE-FIRST VOICE BOX

Thousands of languages can be indexed. Spoken output is limited to voices actually installed on your device. Text and recordings stay on your device in this app, and remote voice services are excluded.

0Local voices detected
0Cloud speech API calls in app
24Linked Living Ears acoustic tracks
384Available Living Eyes views

MICRO · On-device text to speech

Uses a system voice only when the browser identifies it as a localService voice. Missing dialects are shown as missing, never silently invented or routed to the cloud.

Local-only dictation · optional microphone

This feature is disabled unless the browser supports a provable processLocally=true mode and the specified language pack is installed. It will never fall back to remote transcription.

No microphone permission requested.

Family / ancestral recordings · permissions-first

Listen to a recording that you possess and are authorized to use. This plays the original local audio. It does not claim to recreate a grandmother, child, public figure, ancestor or deceased person's voice, and it doesn't clone anyone.

No recording loaded or stored.

Noise reduction · device processing

Choose an audio file to derive a new, quieter version using a local amplitude gate. Original bytes are untouched. This is not neural denoising and may reduce soft syllables.

No audio analyzed.

MACRO · Physical sound and storytelling

Original synthetic sounds inspired by Earth, water, wind and imagined settings. No infrasonic anxiety/disorientation effects, covert manipulation or unsafe playback claims. This is a separate soundboard, not a real recording of Earth's core.

Silent until you start.

MICRO · Actual offline native model runtime

Run the optional local-only Python server on your own computer to connect an independently installed eSpeak NG or whisper.cpp. This section operates only from http://127.0.0.1; the publicly hosted app never sends your content to localhost or a cloud provider.

Not connected. Install local engines and launch the local server first.

Original selected audio remains unchanged. The browser decodes locally, mixes channels, resamples to mono PCM16 at 16 kHz, then passes up to 30 seconds only to the loopback process. A configured local model is required; nothing is downloaded automatically.

Record your own voice · foreground only

Microphone permission is requested only when you press Start. Recording lasts at most 30 seconds, stops when the page is hidden, and stays in temporary page memory unless you explicitly download it. This is your actual recording, not a synthetic clone.

Microphone is off. No capture or background recording.

Device footprint · tokens / privacy / offline

0UTF-8 bytes of current working text
0Outbound audio bytes initiated by app
0Cloud model tokens in app
0Automatic model downloads

App shell cache is opt-in; local OS voice and recognition packs are not part of the HTML bundle.

Native/open-source upgrade routes: sherpa-onnx (offline ASR/TTS), whisper.cpp (ASR), and eSpeak NG (offline CLI). None is bundled or silently installed. The executable desktop eSpeak NG companion is supplied in native/offline_tts.py; model files and licenses are selected separately.