Dialogue Isolate [ADV]
Overview
Dialogue Isolate is designed to separate spoken dialogue from non-stationary background noise such as crowds, traffic, footsteps, weather, or other noise with highly variable characteristics. It can be particularly effective at increasing the level of dialogue in challenging low signal to noise ratio conditions.
Machine learning in Dialogue Isolate
Dialogue Isolate uses a deep neural network, which was trained on large amounts of speech and noise data to automatically recognize the percentage of speech in every time-frequency bin of the spectrogram. Once trained, the neural network processes the incoming audio into separated speech and noise components with independently controllable levels.
Controls
- DIALOGUE GAIN [dB]: Controls the gain of the components in your audio recognized as speech. Leave this slider at 0 dB to reduce noise, or cut to reduce the level of spoken dialogue.
- NOISE GAIN [dB]: Controls the gain of the components in your audio recognized as noise. Keep this slider low to increase dialogue intelligibility, or increase to 0dB while turning down dialogue gain to hear only the isolated noise.
- SEPARATION STRENGTH: When using higher values, the processing will more strictly define what it classifies as dialogue, which can result in more background noise reduction at the cost of possible reduction of speech. When using lower values, the processing will more broadly define what it classifies as dialogue, which will allow more background noise through, but will reduce the possibility of speech loss as a result of processing.
Note
- Dialogue Isolate will still process even when separation strength is set to zero.
- Dialogue Isolate will still process even when separation strength is set to zero.
Alternatives
For stationary noise, such as hiss, buzz, line noise, etc., Dialogue Isolate may produce satisfactory results, but we also suggest trying the Spectral De-noise module in these situations.