Choose the audio task before choosing a microphone: live listening, a conversation with a visitor, sound saved beside video, or an exported clip that another reviewer can hear. These are four separate acceptance requirements. A working talk button does not prove that the NVR records either side of the conversation.
Specify the output you need
| Function | Required path | Proof to request |
|---|---|---|
| Live listening | Site microphone → camera → viewer | Speech audible on the intended phone or workstation |
| Two-way talk | Client microphone → site speaker, plus the listening path | Both people can communicate at normal distance and noise |
| Recorded audio | Camera audio stream → selected recorder channel → saved clip | Playback contains the required sound, not just live video |
| Export | Saved clip → exported file → receiving player | Audio track and timing survive the export workflow |
Ask whether talk is half duplex, where the users take turns, or full duplex, where both can speak at once. If a visitor speaks while an operator holds a talk button, a half-duplex workflow may miss that reply. Also ask whether operator speech is recorded; the return audio path may be different from the camera's microphone stream.
Trace sound through each device
- Microphone input: establish whether a microphone is built in or external. For an external unit, match the input type, connector, required power and permitted level to the camera's manual. A line input and a microphone input are not interchangeable specifications.
- Camera encoding: list the audio codec, sample rate, channel count and bitrate offered by the exact model. H.264 or H.265 describes the video encoding and does not establish which audio format is present.
- Stream selection: identify the stream the NVR actually receives. Test it with audio enabled; hearing sound in the manufacturer's app does not establish that the recorder receives the same stream.
- Recorder storage: verify audio ingestion, per-channel enablement, recording schedule and playback permissions. Check continuous and event recording separately if both are used.
- Client and export: test the actual recorder client, mobile viewer and receiving player. Record the export format and whether it contains an audio track; do not accept a silent file as evidence that the microphone failed.
Axis's Audio API documentation illustrates separate input/output capabilities, duplex modes and product-dependent audio formats including G.711, AAC and Opus. Those are examples of specifications to check, not QuarkView capability claims. ONVIF Profile T covers bidirectional audio for devices and clients supporting that feature. Confirm the exact models and supported functions, then test the combination; an ONVIF logo alone does not settle recording or export.
Make speech intelligible before increasing gain
A controlled intercom point is a more predictable task than a microphone expected to cover a large open yard. Put the pickup near the normal speaking position, away from a fan, vibrating mount or loudspeaker, and use suitable outdoor protection where required. Check the external microphone's environmental rating as well as the camera's rating.
| Symptom | Check first | Repeatable correction test |
|---|---|---|
| Speech buried in noise | Distance, wind, machinery and competing voices | Repeat the same phrase closer to the microphone with the same background noise |
| Harsh or distorted peaks | Input type and excessive gain | Reduce gain and compare normal speech and a loud phrase |
| Echo during talk | Speaker-to-microphone coupling and duplex settings | Test speaker level, position and alternating speech |
| Live sound but silent playback | Recorder stream, codec support, schedule and channel setting | Record a new marked clip after changing one setting |
| Playback works but export is silent | Export options, file audio track and player support | Open the same exported file on the agreed receiving software |
| Broken words or excessive delay | Network path and client performance | Compare local and remote sessions at a busy period |
Increasing gain raises unwanted sound too. Use a listener who does not know the test phrase to judge intelligibility; an operator who knows what was said can overestimate recording quality. A sound-event alarm, if offered, is another feature and needs its own false-alarm test with the site's normal noise.
Example: one entrance conversation point
Planning example, not a tested installation: assume one sheltered entrance, a visitor speaking 1 m from the microphone, one remote operator and a requirement to save site-side speech with entrance video. Specify one verified audio-capable endpoint, a compatible NVR channel and the operator client; keep unrelated corridor microphones disabled. Require the supplier to show whether the operator's voice is also saved if that is needed. The 1 m position is a trial assumption, not a guaranteed pickup range.
Storage example: assume one continuously recorded mono audio stream at 64 kbit/s for 30 days. Audio payload is 64,000 bit/s × 86,400 s/day × 30 days ÷ 8 = 20,736,000,000 bytes, or about 20.7 GB using decimal GB. Four identical channels would add about 82.9 GB. This excludes video, container overhead and reserved capacity; a variable-bitrate stream or a different recording schedule changes the result. Use the configured bitrate, not this example, in the final retention calculation.
Run and retain a complete acceptance record
- Record camera, NVR and client models, firmware versions, stream settings, microphone position, distance and export format. Agree which side of a conversation must be recorded.
- Use a numbered script at normal and louder speech levels. Test normal background activity and an outdoor wind condition where applicable. Keep the distance and conditions with each clip.
- Test listening and talk in both directions. Note turn-taking behavior, interruptions and operator-observed delay; agree tolerances for the actual task.
- Play the saved clip on the NVR and intended remote client. Check the words and timing against a visible action, such as the speaker raising a hand.
- Export the clip and have a second reviewer open it with the receiving software. Verify the audio track, intelligibility and synchronization there.
- Disable capture in the intended control and repeat the recording and live-client checks. Test a user who must not have audio access; verify the actual permission behavior rather than assuming video permissions separate it.
Retest after a firmware update, microphone replacement, changed codec, new export client or moved speaker. Keep acceptance clips under the same access and deletion rules as other recordings.
Define why, where and when sound is captured
For UK organisations, the ICO CCTV checklist says conversation recording is difficult to justify, requires a particular need, and should be disclosed; it suggests switching audio on only in specific circumstances, and addresses notices, limited retention and secure access. This is UK organisational guidance; it does not establish conversation-recording rules for another country or state.
Translate the permitted purpose into settings: selected channels and hours, authorised listening/playback/export users, a justified retention period and a named person to review changes. A video privacy mask should not be assumed to mute a microphone. Reception, workplace and intercom use require the applicable rules for that setting to be checked before recording. Avoid broad pickup that captures unrelated confidential conversations when a focused communication point can meet the task.
Send the speaking position, noise conditions, required audio functions, existing recorder and export workflow with your configuration inquiry. Use the camera/NVR compatibility checklist and two-way audio guide for the adjoining decisions.