RADIANCE VOICE LAB

Choose a talkgroup and press Play DMR.

Connecting to live talkgroup activity…

Stopped.

Hoseline fallback ↗
Choose a passage · up to 60 seconds

Translation

Enable translation once to download the models. Audio stays on this device.

Fallback · share tab audio

Native DMR audio feeds translation while playing. If the receiver is unavailable, use Hoseline and tab-audio capture below.

Tab audio unavailable?

Some browsers do not supply tab audio. If sharing returns no audio, use a recording or select a radio audio input above. Selecting a talkgroup is not audio capture.

Not capturing. No microphone is selected automatically.

Advanced & offline tools
Offline settings

Not checked on this device.

Diagnostics
Model
not-prepared
Audio source
none · stopped
Audio level
No audio measured
Last processing time
No completed passage
Backlog
0 pending · 0 skipped
Recent errors
No recorded errors

Measurements only. No audio, transcript, filename or notebook content.

Built-in self-test

A bundled English speech excerpt tests transcription and Spanish translation. First run may download models. Failed word comparisons remain visible.

Not run. Cache-only processing does not prove offline site reopening.

    Sample source and reference

    Offline setup and test limits

    The first preparation downloads the multilingual Whisper Base model, which needs more download time and device memory than the earlier Tiny model; Spanish or French output adds a separate translation model download. The app and model are cached in this browser. Reopen this page offline to verify readiness before field use. Browser storage can be cleared or evicted.

    Start with a 10–30 second recording. Maximum selected passage: 60 seconds. Maximum uploaded file: 100 MB. File processing uses a selected passage; live tab audio processes bounded 15-second passages and may lag or skip audio if the computer cannot keep up. Recording captures the browser’s default microphone or audio input, not computer output. Select a radio USB input in your device settings if needed.

    No audio or transcript is uploaded. Download requests go to Hugging Face. Machine text can invent words in noise; verify callsigns against the recording. Not an emergency communications tool.

    Audio controls ↑

    Heard · running transcript

    Understood · running translation

    English

    Replay & review

    Every machine passage needs checking. No calibrated confidence score is available.

    Radio vocabulary & callsigns

    Review aids, not model training. Exact matches are highlighted below; near matches are suggestions only. Original machine text stays intact.

    Test notebook

    Notebook stays on this browser. Export a backup.