espnet2.gan_tts.melgan.pqmf.PQMF

source

class espnet2.gan_tts.melgan.pqmf.PQMF(subbands: int = 4, taps: int = 62, cutoff_ratio: float = 0.142, beta: float = 9.0)

Bases: Module

Pseudo QMF modules.

This code is modified from https://github.com/kan-bayashi/ParallelWaveGAN.

The PQMF (Pseudo Quadrature Mirror Filter) module is designed for near-perfect reconstruction filter banks based on the concept of pseudo-QMF banks. It includes methods for both analysis and synthesis of signals using these filter banks.

The implementation uses a Kaiser window approach to design the prototype filters that are fundamental to the PQMF operation.

analysis_filter

Analysis filter coefficients.

• Type: torch.Tensor

synthesis_filter

Synthesis filter coefficients.

• Type: torch.Tensor

updown_filter

Filter for downsampling and upsampling.

• Type: torch.Tensor

subbands

Number of subbands in the PQMF.

• Type: int

• Parameters:

  • subbands (int) – The number of subbands.
  • taps (int) – The number of filter taps.
  • cutoff_ratio (float) – Cut-off frequency ratio.
  • beta (float) – Beta coefficient for Kaiser window.

analysis(x

torch.Tensor) -> torch.Tensor: Applies the analysis operation on the input tensor using PQMF.

synthesis(x

torch.Tensor) -> torch.Tensor: Applies the synthesis operation on the input tensor using PQMF.

Examples

>>> pqmf = PQMF(subbands=4, taps=62, cutoff_ratio=0.142, beta=9.0)
>>> input_tensor = torch.randn(1, 1, 1024)  # Example input
>>> analysis_output = pqmf.analysis(input_tensor)
>>> synthesis_output = pqmf.synthesis(analysis_output)

Notes

• The cutoff_ratio and beta parameters are optimized for subbands = 4. See discussion in https://github.com/kan-bayashi/ParallelWaveGAN/issues/195.
• Power will be decreased in the synthesis process, so the output is multiplied by the number of subbands. This approach should be reviewed for correctness.
• Further understanding of the reconstruction procedure is needed (TODO).

Initilize PQMF module.

The cutoff_ratio and beta parameters are optimized for #subbands = 4. See dicussion in https://github.com/kan-bayashi/ParallelWaveGAN/issues/195.

• Parameters:
  • subbands (int) – The number of subbands.
  • taps (int) – The number of filter taps.
  • cutoff_ratio (float) – Cut-off frequency ratio.
  • beta (float) – Beta coefficient for kaiser window.

#

analysis(x

: Tensor) → Tensor

Pseudo QMF modules.

This code is modified from https://github.com/kan-bayashi/ParallelWaveGAN.

#

synthesis(x

: Tensor) → Tensor

Pseudo QMF modules.

This code is modified from https://github.com/kan-bayashi/ParallelWaveGAN.