Skip to main content
Background noise on the caller’s end — traffic, crowds, wind, TV, children — degrades transcription accuracy and causes the agent to mishear or miss intent. DialNexa applies server-side denoising to the caller’s audio stream before it reaches the transcriber, reducing noise without requiring any client-side changes. Denoising happens on the incoming audio only. It does not affect the agent’s outgoing TTS audio.

​
Denoising Mode

Set Denoising Mode in your agent’s Speech Settings under the Input Audio section.
ModeBehaviorUse when
OffNo denoising applied. Raw caller audio goes to the transcriber.Callers are in controlled environments (call centers, quiet offices). Noise filtering is not needed.
LowLight noise suppression. Removes low-level hiss and mild background hum.Light background noise is common but callers are generally in quiet spaces. Minimal risk of speech distortion.
HighAggressive noise suppression. Removes most background noise including loud environments.Callers are frequently outdoors, in cars, or in noisy households.
High denoising mode may reduce transcription accuracy for callers with non-standard accents or speech patterns. The noise removal algorithm occasionally attenuates certain phonemes alongside ambient noise. Test with real callers from your target population before deploying High mode widely.

​
Interaction with transcription accuracy

Denoising improves transcription in most noisy environments, but the relationship is not linear:
  • Quiet environments + High denoising: unnecessary processing, potential slight degradation of clear speech
  • Moderate noise + Low denoising: best balance — suppresses noise without touching speech characteristics
  • Very noisy environments + High denoising: large transcription accuracy improvement; some accuracy loss possible for accented speech
  • Very noisy environments + Off: poor transcription, high word error rate
For callers with strong regional accents (Indian English, Arabic-accented English, etc.), start with Low denoising and test before switching to High.

​
Testing noise handling

Before deploying an agent to live callers, test denoising with representative audio conditions.
1

Make a test call in a noisy environment

Use a real phone (not a VoIP softphone in ideal conditions) from a location that matches your expected caller environment — a car, a street, or a busy indoor space.
2

Review the transcript in Session History

Open Monitor > Session History, find the test call, and inspect the transcription events. Look for substitution errors (wrong words), deletions (words missed entirely), and hallucinated words.
3

Adjust Denoising Mode

If word errors are high and the environment was noisy, increase denoising. If the caller had an accent and High mode was active, try Low.
4

Repeat

Run at least 3 to 5 test calls per denoising level before concluding. Noise conditions vary call to call.

​
Audio Cache interaction

Denoising applies to incoming caller audio and does not interact with the Audio Cache, which is an outgoing TTS cache. These two settings are independent.