RoomImpulseResponseOptions

Options & Configuration · AiDotNet.Audio.Effects

Configuration options for the FiNS (Filtered Noise Shaping) room impulse response estimator.

Inheritance: ModelOptions → RoomImpulseResponseOptions

For Beginners

When you clap in a room, you hear a sharp initial sound followed by a smeared tail of echoes. Those two parts behave very differently: the beginning has clear structure worth predicting exactly, while the tail is essentially noise that fades out at a different speed in each frequency band. FiNS mirrors that split — it predicts the sharp start directly, and it makes the tail by taking noise, splitting it into frequency bands, and learning how loud each band should be over time. Modelling the tail as shaped noise rather than predicting 48,000 individual samples is what makes the problem tractable.

How It Works

FiNS estimates a time-domain room impulse response directly from reverberant speech. A strided 1-D convolutional encoder compresses the waveform to a single latent embedding, and a decoder then builds the response as two parts: a short early component predicted sample-by-sample, and a late field synthesised by shaping filtered noise with predicted time-domain masks.

Reference: Steinmetz, Ithapu and Calamia, “Filtered Noise Shaping for Time Domain Room Impulse Response Estimation from Reverberant Speech”, WASPAA 2021 (arXiv:2107.07503). Defaults below reproduce the paper’s configuration; each is annotated with its source.

Constructors

Constructor Summary
RoomImpulseResponseOptions()

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } Gets or sets the ONNX runtime options.
double DereverberationStrength { get; set; } Gets or sets the dereverberation strength (0-1).
double LearningRate { get; set; } Gets or sets the learning rate.
double RT60WindowSeconds { get; set; } Gets or sets the RT60 estimation window in seconds.
int EarlyResponseLength { get; set; } Gets or sets the early-component length E in samples.
int EncoderKernelSize { get; set; } Gets or sets the encoder convolution kernel size.
int EncoderMaxChannels { get; set; } Gets or sets the channel count reached at the final encoder block.
int EncoderStride { get; set; } Gets or sets the encoder convolution stride.
int LatentDim { get; set; } Gets or sets the dimensionality of the latent embedding z.
int NoiseFilterOrder { get; set; } Gets or sets the FIR order P of each learned band-pass filter.
int NumDecoderBlocks { get; set; } Gets or sets the number of decoder blocks.
int NumEncoderBlocks { get; set; } Gets or sets the number of strided convolutional encoder blocks.
int NumNoiseBands { get; set; } Gets or sets the number of noise bands M.
int RIRLength { get; set; } Gets or sets the estimated RIR length L in samples.
int SampleRate { get; set; } Gets or sets the audio sample rate in Hz.
int[] StftFrameSizes { get; set; } Gets or sets the STFT frame sizes for the multi-resolution STFT loss.
string ModelPath { get; set; } Gets or sets the path to the ONNX model file.

Methods

Method Summary
void Validate() Validates that the configuration describes a buildable FiNS model.