NeuroDMR Monitor: DMR Monitoring with AI — Live Subtitles & Translation

NeuroDMR Monitor listening to BrandMeister TG 91 (World-wide): the decoded audio as a spectrum and waterfall, with live English subtitles translated into Japanese.

TEST BUILD — A work-in-progress demo by Ian Lee / KD8CEC. Features and settings may change.

Executable release planned for this week — The Windows build of NeuroDMR Monitor will be posted here at hamskey.com under the NeuroDMR label. No download link yet.

NeuroDMR Monitor: DMR Monitoring with AI — Live Subtitles & Translation

NeuroDMR Monitor is a desktop app I am building to read DMR traffic as well as hear it. It listens to BrandMeister talkgroups, turns each transmission into text with speech recognition, and translates it into the language of your choice — live, while the operators are still talking. In this recording I hop across talkgroups in Germany, the world-wide TG 91, Malaysia and France, and change the target language several times along the way.

How it fits together

The pipeline as configured in this demo. The LLM layer is configured but not exercised in this recording.

Audio comes from the BrandMeister Hoseline, so no account or radio is needed just to listen. A hotspot connection with local AMBE decoding is offered as an alternative source. The talkgroup map and activity list are fed by the BrandMeister Last Heard live stream.

The settings used in this demo

Speech recognition: Groq Whisper with whisper-large-v3-turbo, automatic language detection with a language hint for national talkgroups, and realtime updates every 6 seconds. (API key masked.)

Other engines are available in the same menu: a built-in Whisper that runs on the PC without a key (the default), OpenAI Whisper, and a local or self-hosted Whisper server. Translation in this session uses Google, which needs no key.

The AI (LLM) tab: provider Groq, model openai/gpt-oss-120b. “AI analysis after every call” is switched off for this recording.

Changing languages on the fly

On TG 264888 (Germany) the German conversation is detected automatically and translated into English. While listening I switch the target to Swedish and then to Japanese. The change applies to new text: a transmission that was already in progress keeps its earlier translation and continues in the new language.

Switching from Swedish to Japanese mid-transmission: the earlier part stays Swedish, the new text is Japanese.

Tags and keyword alerts

Some lines receive small tags — callsigns and common ham-radio expressions such as “73 = best regards” or “CQ = calling any station”. “CQ” is also one of the alert keywords in this setup (with “emergency” and “mayday”), so the line is highlighted and a notification appears.

A “CQ” call on TG 91 is highlighted with a keyword alert. Note the phonetic miss: “Foxtrot Yankee Romeo” was heard as “Foxtrot Thank You Romeo”.

What worked

• Language detection picked up German, English, Malay and French without manual selection.
• Conversational speech — for example casual weather talk on TG 91 — was transcribed cleanly and translated.
• The target language can be changed while listening; German, English and French were translated into English, Swedish, Japanese and Chinese during the session.
• A callsign tag corrected itself as more audio arrived, and the final English line “20 dBm is 100 mW” was recognized and translated correctly.

French technical talk on TG 20854. An earlier attempt heard “20 dBm” as “20 BB”; the later line “20 dBm, c'est 100 mW” → “20 dBm is 100 mW” is correct.

What to improve

• Ham-radio vocabulary. Terms such as FreeDV and PSK Reporter came out as “FriedeV” and “PSK-Deporter”, and phonetic alphabet letters can be misheard.
• Literal translations. Misheard or ambiguous words are translated literally — “Echse” became “the lizard”, and the radio sense of “mode” became “fashion”.
• Units and numbers. “20 dBm” was first recognized as “20 BB”, which made that sentence wrong.
• Short, repetitive transmissions (the Malay talkgroup in this test) are hard to judge from the screen alone.

Natural next steps are a ham-radio vocabulary hint for recognition and, optionally, the per-call LLM correction that was left off in this recording. I have not tested yet how much either one helps.

The executable is planned for release this week; watch the NeuroDMR label for the post.

Quick Links: NeuroDMR posts · NeuroSDR projects

73,
Ian Lee / KD8CEC