Skip to main content

Aligner - PyTorch

Project description

Aligner - PyTorch

Sequence alignement methods with helpers for PyTorch.

Install

pip install aligner-pytorch

PyPI - Python Version

Usage

MAS

MAS (Monotonic Alignment Search) from GlowTTS. This can be used to get the alignment of any (similarity) matrix. Implementation in optimized Cython.

from aligner_pytorch import mas

sim = torch.rand(1, 4, 6) # [batch_size, x_length, y_length]
alignment = mas(sim)

"""
sim = tensor([[
    [0.2, 0.8, 0.9, 0.9, 0.9, 0.4],
    [0.6, 0.8, 0.9, 0.7, 0.1, 0.4],
    [1.0, 0.4, 0.4, 0.2, 1.0, 0.7],
    [0.1, 0.3, 0.1, 0.7, 0.6, 0.9]
]])

alignment = tensor([[
    [1, 0, 0, 0, 0, 0],
    [0, 1, 1, 1, 0, 0],
    [0, 0, 0, 0, 1, 0],
    [0, 0, 0, 0, 0, 1]
]], dtype=torch.int32)
"""

XY Embedding to Alignment

Used during training to get the alignement of a x_embedding with y_embedding, computes the log probability from a normal distribution and the alignment with MAS.

from aligner_pytorch import get_alignment_from_embeddings

x_embedding = torch.randn(1, 4, 10)
y_embedding = torch.randn(1, 6, 10)

alignment = get_alignment_from_embeddings(
    x_embedding=torch.randn(1, 4, 10),  # [batch_size, x_length, features]
    y_embedding=torch.randn(1, 6, 10),  # [batch_size, y_length, features]
)                                       # [batch_size, x_length, y_length]

"""
alignment = tensor([[
    [1, 0, 0, 0, 0, 0],
    [0, 1, 0, 0, 0, 0],
    [0, 0, 1, 0, 0, 0],
    [0, 0, 0, 1, 1, 1]
]], dtype=torch.int32)
"""

Duration Embedding to Alignment

Used during inference to compute the alignment from a trained duration embedding.

from aligner_pytorch import get_alignment_from_duration_embedding

alignment = get_alignment_from_duration_embedding(
    embedding=torch.randn(1, 5),    # Embedding: [batch_size, x_length]
    scale=1.0,                      # Duration scale
    y_length=10                     # (Optional) fixes maximum output y_length
)                                   # Output alignment [batch_size, x_length, y_length]

"""
alignment  = tensor([[
    [1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
    [0, 0, 0, 1, 0, 0, 0, 0, 0, 0],
    [0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
    [0, 0, 0, 0, 0, 1, 1, 1, 0, 0],
    [0, 0, 0, 0, 0, 0, 0, 0, 1, 0]
]])
"""

Citations

Monotonic Alignment Search

@misc{2005.11129,
Author = {Jaehyeon Kim and Sungwon Kim and Jungil Kong and Sungroh Yoon},
Title = {Glow-TTS: A Generative Flow for Text-to-Speech via Monotonic Alignment Search},
Year = {2020},
Eprint = {arXiv:2005.11129},
}

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

aligner-pytorch-0.0.19.tar.gz (108.5 kB view details)

Uploaded Source

File details

Details for the file aligner-pytorch-0.0.19.tar.gz.

File metadata

  • Download URL: aligner-pytorch-0.0.19.tar.gz
  • Upload date:
  • Size: 108.5 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.1 CPython/3.9.15

File hashes

Hashes for aligner-pytorch-0.0.19.tar.gz
Algorithm Hash digest
SHA256 0a7e41b3586aa062a91c8c3c0346275378fa896a96d81315c2b21d0c3d678804
MD5 618f989d74ab8e70fce9a31789fae9d4
BLAKE2b-256 6ec32aa2dce18b99182a36df4fa8c24f2df72bb1da03c16374f6c1b368bc839b

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page