From audio to verdict
A voice note or video circle arrives
Telm picks up voice messages and round video notes automatically — no commands, no manual review queue.
The audio becomes text
The recording is transcribed in any of 25 languages. From here on, spoken spam is just spam.
The same layered checks run
The transcript goes through blocklists, 150+ rules, our own ML model — and an LLM only for borderline cases. Exactly the pipeline that checks every written message.
The decision lands in the journal
The transcript is saved next to the verdict, so nothing is a black box — and the author can appeal, same as for any text message.
Built to be trusted, not just obeyed
You choose whose audio is checked
Check new members only — the usual source of voice spam — or everyone in the group.
Monitoring mode first
Voice moderation starts by showing you the decisions it would make. Enforcement switches on only when you're ready.
25 languages, zero setup
Transcription detects the spoken language automatically — a pitch in your community's language is understood without configuration.
Genuine voice notes pass through
Only spoken spam is acted on. A transcript is scored by the whole pipeline, so a single mis-heard word can't punish a real member.
How spoken spam gets caught
When a voice note or a video circle arrives, Telm transcribes the audio — it understands 25 languages, with nothing to configure — and from that moment the recording is just text. The transcript runs through the same layered pipeline that guards your group against written spam: blocklists first, then 150+ rules, then our own ML model, and an LLM only for the borderline cases. If the spoken content is spam, it's handled exactly like a written message would be, and the transcript is recorded in the moderation journal next to the decision, so you can read precisely why a clip was removed. Members keep the same rights too: any action on a voice message can be appealed, just like any other.
You stay in control of how far it reaches. Like the rest of Telm, voice moderation starts in monitoring mode — you watch the decisions it would make before switching on enforcement. You also choose whose audio gets checked; new members are the usual source of voice spam, so many groups start there. The honest caveat: no transcription is perfect. That's exactly why a transcript is scored by the full pipeline rather than punished on a single matched word — a mumbled phrase can't ban anyone, and genuine voice notes pass through untouched.
Frequently asked questions
What kinds of audio does Telm check?
Voice messages and round video circles. Both are transcribed automatically and the recognized text runs through the same anti-spam checks as a written message.
Which languages does it understand?
Transcription supports 25 languages and detects the spoken language automatically — there is nothing to configure per group.
Can a bad transcription get someone banned?
No. The transcript is scored by the full layered pipeline — blocklists, 150+ rules, our own ML, and an LLM for borderline cases — so no single mis-heard word can trigger a punishment. And every action can be appealed, just like for text.
What happens when spam is found in a voice note?
The same thing that happens with written spam: the message is handled by your group's settings, the decision is recorded in the moderation journal together with the transcript, and the author can appeal it.
Do I have to trust it from day one?
No. Voice moderation works in monitoring mode first: you see every decision it would make without anything being enforced, and you switch on enforcement when the calls look right for your group.
Related features
Explore more ways Telm can help your community