IPSC3 Parrot calls do more than replay transmitted audio. After the recording is played back, Parrot identifies the caller and announces a decoded-audio report. The same test is retained in the dashboard with additional technical measurements that can be reviewed later.
The practical target is Level normal with no clipping detected. This indicates useful speech level without repeatedly driving the decoded audio close to full scale. Avoid using a radio that repeatedly reports a high level or clipping, because excessively hot audio usually becomes harsh, distorted, and less intelligible after AMBE encoding.
A typical spoken report is:
Level normal, no clipping detected, peak amplitude 72 percent,
average RMS 16 percent.The spoken report deliberately gives the most useful results first. Detailed dBFS, sample, duration, and decoder information remains available on the dashboard without making the radio announcement unnecessarily long.
Active speech is below the empirical network reference range. Listen to the replay first: a low report is advisory, not proof of an incorrectly adjusted radio. Confirm the microphone and accessory settings and repeat with normal speech at your usual distance. Only increase gain if the replay is genuinely too quiet; do not spoil clear on-air audio just to change the label.
Active speech is in the preferred range. If the replay is intelligible and Parrot also reports no clipping, no adjustment is normally required. Do not increase microphone gain merely to obtain a larger percentage.
Active speech is too hot, or enough samples are close to full scale to trigger the clipping limit. Reduce radio or accessory microphone gain, avoid speaking directly into the microphone opening, and do not shout. Repeat the test until a representative transmission reports normal without clipping.
Clipping means a significant portion of the decoded waveform is repeatedly reaching the near-full-scale region. This removes speech detail and can make AMBE audio sound harsh, rough, or difficult to understand. A small number of isolated peaks can appear in the detailed measurements, but the spoken clipping detected warning is issued when at least 0.5 percent of decoded samples are near full scale. Lower the microphone gain until the warning clears.
Peak amplitude is the highest instantaneous decoded PCM level seen anywhere in the test, expressed as a percentage of full scale. It is useful for finding sharp peaks, but it does not describe the average loudness of speech. A high peak without clipping may be acceptable; use the overall level, clipping result, and replay quality together.
Average RMS represents the effective level of active speech rather than one brief peak. Silence below the speech gate is excluded, making this the main measurement used to classify microphone level. It is shown as a percentage in the spoken result and as both a percentage and dBFS on the dashboard.
Audio Reports comes directly from the retained Parrot report table. It does not depend on an ordinary call-history row being present, and ordinary voice calls are not analysed.
The local date and time when analysis completed. Use this to compare a report with a specific radio or CPS microphone-gain change.
The transmitted DMR Radio ID and the callsign and name resolved from the Radio ID database. If database details are unavailable, the numeric Radio ID remains the authoritative source identity.
The protocol and endpoint through which the test entered IPSC3, such as an MMDVM hotspot, Motorola IPSC repeater, or POC endpoint. This helps distinguish tests made through different radios or access paths.
The overall low, normal, or high classification. In analysis version 3, active-speech RMS below -35 dBFS is low, -35 through -13 dBFS inclusive is normal, and above -13 dBFS is high. A report is also high when clipping reaches the configured 0.5 percent threshold. Use this as a guide alongside replay intelligibility and on-air reports, not as a calibrated transmitter alignment.
The lower boundary was revised after many good-sounding radios were reported low. The review covered 589 retained tests from 16 to 23 September 2026 on the VK server. To reduce short-test and repeated-user bias, the reference used 134 unclipped, non-high tests with at least five seconds of audio and three seconds above the speech gate, then gave each radio ID/access-path median equal weight: 52 combinations across 50 radio IDs. The lower fifth-percentile level was -34.015 dBFS; rounding down to -35 dBFS provides a small margin.
This is an empirical operating reference, not a formal DMR loudness standard or proof that every reference recording was ideal. Version 3 changes only the low/normal boundary. Audio gain, decoding, the speech gate, the high boundary and clipping detection are unchanged. New tests use version 3; historical version 2 tests keep their original labels and -24 dBFS low boundary. No historical measurements or spoken reports are rewritten.
The largest absolute decoded PCM sample in the call. The table shows percent of full scale, dBFS, and the raw sample amplitude. Zero dBFS is digital full scale, so negative values have more headroom. Peak is a momentary maximum and should not be treated as average speech level.
The speech-gated root-mean-square level calculated from active speech samples. The table shows percent and dBFS. Whole RMS also includes quiet portions of the complete recording, so it will normally be lower and is most useful when comparing tests of similar duration and timing.
The percentage and count of decoded samples at or above the near-full-scale threshold. The analyser currently treats absolute sample amplitudes of 31,120 or greater, approximately 95 percent of full scale, as clipped samples. The spoken warning and high classification occur at 0.5 percent clipping. Aim for no clipping rather than accepting persistent high or clipped reports.
These values are measured after AMBE decoding. They are not calibrated RF deviation, microphone voltage, BER, RSSI, or proof that every aspect of a transmitter is correctly aligned. Codec scaling, radio audio processing, accessories, background noise, and speaking technique can influence the result. Always listen to the replay as well as reading the numbers.
Every standard IPSC3 voice pack contains the same Parrot report vocabulary. New voices and languages must provide these samples through the offline prompt builder before selection, keeping report behaviour consistent without adding speech synthesis or vocoding work to the live packet path.