IPSC3 uses generated AMBE voice prompts across its DMR endpoint types for events such as scan linked/unlinked, dynamic talkgroup linking and timeout, listen-only busy feedback, unavailable private-call feedback, Parrot identity reports, and other network status messages.
The selected IPSC3 voice pack supplies reusable digits, letters, phonetic words, and network feedback words. The daemon assembles the requested phrase from those AMBE assets and then frames it for the destination endpoint.
Administrators select the default generated voice under Configuration > System > Default Signal Voice. Current English packs include Australian voices Matt, Alex, Amelia and Her; New Zealand voice Belinda; US voice Jim; and UK voices Steve and Charlotte. Changing the selection restarts the IPSC3 radio daemon so the new AMBE pack is loaded consistently.
On 23 September 2026, all eight voices were rebuilt through a physical AMBE3000R and installed on VK, ZL, UK, US and FreeDMR-UK. The library contains 928 recordings and 37,051 verified AMBE frames. Each site's selected voice was preserved and rollback copies retained. No radio daemon was restarted during the asset rollout; the POC Parrot cache updates at the next controlled IPSC3 restart.
Voice packs can be built offline using a physical AMBE3000 or AMBE3003 encoder from the original normalized speech recordings. Existing AMBE recordings are not decoded and re-encoded. The words, phonetic alphabet, Parrot report vocabulary and callback phrases remain the same.
Each sample starts with fresh encoder state. The completed pack retains its voice identity and exact symbol index, and is checked for frame order, duration, hashes and zero error-correction changes. Hardware-only builds never substitute software-encoded samples. A failed sample is discarded and may be restarted in full; a damaged partial sample is not published.
The resulting AMBE files are ordinary reusable assets: other IPSC3 servers do not need a hardware dongle to play them. Distribution preserves each server's selected voice. Normal DMR announcements read the installed files; POC Parrot uses a startup cache and needs a controlled IPSC3 restart after replacing that pack. Native PCM broadcast and callback audio is unchanged. Decoder level measurements are retained for review rather than treated as a substitute for listening tests on real radios.
Administrators can enable Invalid Radio ID Voice Warning in Configuration > System > Core Options. It is disabled by default. When enabled, IPSC3 lets a voice transmission complete normally and then checks the source radio ID against the local RadioID database. If the ID is unknown, IPSC3 returns a private voice announcement to that user: Your radio ID is invalid. Please check your radio ID.
MMDVM prompt playback uses the endpoint templates learned from its normal DMR traffic. IPSC3 sends a valid voice header, the timed AMBE prompt bursts, and a voice terminator. Simplex hotspots use their effective single RF timeslot; duplex hotspots and provisioned MMDVM repeaters use the requested timeslot.
Motorola repeaters and subscriber radios are stricter than many MMDVM endpoints about stream headers, Link Control identity, and the first voice burst. IPSC3 therefore sends prompts as complete Motorola group-voice streams rather than generic AMBE bursts.
Hytera prompts use the shared DMR AMBE assets and timing. IPSC3 builds the DMR stream from learned templates and translates it through the Hytera adapter for the target repeater.
Dashboard configuration provides the general prompt pre-roll and tail-hold settings. The encoded one-second silence is additional in-band protection against first-syllable clipping. Motorola playback also enforces its minimum framing pre-roll.
linked, unlinked, and now make first-syllable clipping easiest to hear.