ComfyUI Node: Egregora DeepFilterNet Denoise

Egregora DeepFilterNet Denoise:

Egregora_DeepFilterNet_Denoise is a sophisticated audio processing node designed to enhance speech quality by reducing noise using the DeepFilterNet technology. Operating at a native rate of 48 kHz, this node is capable of handling both mono and stereo audio inputs, offering flexibility in processing either per-channel or by downmixing to mono. It incorporates a 'strength' parameter that allows you to control the wet/dry mix using either an equal-power or linear curve, ensuring a balanced output. The node also features an adaptive mix driven by Voice Activity Detection (VAD), which can utilize RNNoise if available, or an energy/RMS proxy otherwise. Additionally, it includes a post-gain adjustment and a simple peak ceiling limiter to maintain audio quality and prevent distortion. This node is particularly beneficial for enhancing speech clarity in noisy environments, making it an essential tool for audio enhancement tasks.

Egregora DeepFilterNet Denoise Input Parameters:

audio

This parameter represents the audio input that you wish to process. It can be either mono or stereo, and the node will handle it accordingly based on the stereo_mode setting.

dfn_model

Specifies the version of the DeepFilterNet model to use, with "DeepFilterNet2" as the default option. This choice affects the underlying algorithm used for noise reduction.

device

Determines the computational device for processing, with "auto" as the default, allowing the node to select the best available option between CPU and GPU.

use_postfilter

A boolean parameter that, when enabled, applies an additional post-filtering step to further enhance audio quality. The default is False.

limit_ceiling

A boolean parameter that, when enabled, applies a peak ceiling limiter to prevent audio clipping. The default is True.

stereo_mode

Defines how stereo audio is processed, with options like "per_channel" for independent channel processing or downmixing to mono.

frame_ms

Sets the frame size in milliseconds for processing, with a default value of 20 ms. This affects the granularity of the noise reduction.

strength

Controls the wet/dry mix of the processed audio, with a default value of 0.65. A higher value increases the effect of noise reduction.

mix_curve

Determines the curve used for mixing, with options like "equal_power" for a balanced mix.

adaptive_vad_source

Specifies the source for adaptive VAD, with options like "rms" for energy-based detection.

adaptive_mode

Defines the mode of adaptation, with options like "more_on_noise" to increase processing in noisy segments.

adaptive_amount

Controls the degree of adaptation, with a default value of 0.45, affecting how much the VAD influences the mix.

vad_threshold

Sets the threshold for VAD activation, with a default value of 0.90, determining sensitivity to speech presence.

vad_smooth_ms

Specifies the smoothing duration in milliseconds for VAD, with a default of 60 ms, affecting the responsiveness of the VAD.

post_gain_db

Adjusts the post-processing gain in decibels, with a default value of 0.5 dB, allowing for fine-tuning of output volume.

ceiling

Sets the peak ceiling level, with a default value of 0.98, to prevent audio clipping.

Egregora DeepFilterNet Denoise Output Parameters:

processed_audio

The output is the enhanced audio with reduced noise, maintaining the original sample rate and format. This processed audio is clearer and more intelligible, especially in noisy environments.

Egregora DeepFilterNet Denoise Usage Tips:

• Experiment with the strength parameter to find the optimal balance between noise reduction and audio clarity for your specific use case.
• Utilize the adaptive_vad_source and adaptive_mode settings to tailor the node's response to varying noise levels in your audio input.
• Adjust the post_gain_db and ceiling parameters to ensure the output audio maintains a consistent volume without clipping.

Egregora DeepFilterNet Denoise Common Errors and Solutions:

"Invalid audio input format"

• Explanation: The audio input provided is not in a supported format or sample rate.
• Solution: Ensure that the audio input is in a compatible format, such as WAV, and has a sample rate of 48 kHz.

"Device not found"

• Explanation: The specified device for processing is unavailable or incorrectly specified.
• Solution: Set the device parameter to "auto" to allow the node to automatically select the best available device.

"VAD threshold too high"

• Explanation: The vad_threshold is set too high, causing the VAD to miss speech segments.
• Solution: Lower the vad_threshold to a more appropriate level to ensure accurate speech detection.