neural_source_filter package

class neural_source_filter.NSFHifiganGenerator(in_channels: int, out_channels: int, resblock_type: str, resblock_dilation_sizes: Sequence[Sequence[int]], resblock_kernel_sizes: Sequence[int], upsample_kernel_sizes: Sequence[int], upsample_initial_channel: int, upsample_factors: Sequence[int], inference_padding: int = 5, cond_channels: int = 0, conv_pre_weight_norm: bool = True, conv_post_weight_norm: bool = True, conv_post_bias: bool = True, sine_gen: SineGen = SineGen())

Bases: HifiganGenerator

forward(x: torch.Tensor, g: torch.Tensor | None = None, f0: torch.Tensor = tensor([])) → torch.Tensor

Args:

x (Tensor): feature input tensor. g (Tensor): global conditioning input tensor. f0 (Tensor): fundamental frequency input tensor.

Returns:

Tensor: output waveform.

Shapes:

x: [B, C, T] g: [B, C_g, T] f0: [B, 1, T] or [B, T]

training: bool

class neural_source_filter.SineGen(samp_rate: int, harmonic_num: int = 0, sine_amp: float = 0.1, noise_std: float = 0.003, voiced_threshold: float = 0, flag_for_pulse: bool = False)

Bases: Module

forward(f0: Tensor) → tuple[Tensor, Tensor, Tensor]

sine_tensor, uv = forward(f0) input F0: tensor(batchsize=1, length, dim=1)

f0 for unvoiced steps should be 0

output sine_tensor: tensor(batchsize=1, length, dim) output uv: tensor(batchsize=1, length, 1)

training: bool

class neural_source_filter.SourceModuleHnNSF(sine_gen: SineGen, activation: Module = Tanh())

Bases: Module

SourceModule for hn-nsf SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,

add_noise_std=0.003, voiced_threshod=0)

sampling_rate: sampling_rate in Hz harmonic_num: number of harmonic above F0 (default: 0) sine_amp: amplitude of sine source signal (default: 0.1) add_noise_std: std of additive Gaussian noise (default: 0.003)

note that amplitude of noise in unvoiced is decided by sine_amp

voiced_threshold: threshold to set U/V given F0 (default: 0) Sine_source, noise_source = SourceModuleHnNSF(F0_sampled) F0_sampled (batchsize, length, 1) Sine_source (batchsize, length, 1) noise_source (batchsize, length 1) uv (batchsize, length, 1)

forward(f0: Tensor) → tuple[Tensor, Tensor, Tensor]

Sine_source, noise_source = SourceModuleHnNSF(F0_sampled) F0_sampled (batchsize, length, 1) Sine_source (batchsize, length, 1) noise_source (batchsize, length 1)

l_sine_gen: SineGen