Prososdy Morph: A python library for manipulating pitch and duration in an algorithmic way, for resynthesizing speech.

Related tags

Deep LearningProMo
Overview

ProMo (Prosody Morph)

https://travis-ci.org/timmahrt/ProMo.svg?branch=master https://coveralls.io/repos/github/timmahrt/ProMo/badge.svg?branch=master https://img.shields.io/badge/license-MIT-blue.svg?

Questions? Comments? Feedback? Chat with us on gitter!

Join the chat at https://gitter.im/pythonProMo/Lobby

A library for manipulating pitch and duration in an algorithmic way, for resynthesizing speech.

This library can be used to resynthesize pitch in natural speech using pitch contours taken from other speech samples, generated pitch contours, or through algorithmic manipulations of the source pitch contour.

1   Common Use Cases

What can you do with this library?

Apply the pitch or duration from one speech sample to another.

  • alignment happens both in time and in hertz

    • after the morph process, the source pitch points will be at the same absolute pitch and relative time as they are in the target file
    • time is relative to the start and stop time of the interval being considered (e.g. the pitch value at 80% of the duration of the interval). Relative time is used so that the source and target files don't have to be the same length.
    • temporal morphing is a minor effect if the sampling frequency is high but it can be significant when, for example, using a stylized pitch contour with few pitch samples.
  • modifications can be done between entire wav files or between corresponding intervals as specified in a textgrid or other annotation (indicating the boundaries of words, stressed vowels, etc.)

    • the larger the file, the less useful the results are likely to be without using a transcript of some sort
    • the transcripts do not have to match in lexical content, only in the number of intervals (same number of words or phones, etc.)
  • modifications can be scaled (it is possible to generate a wav file with a pitch contour that is 30% or 60% between the source and target contours).

  • can also morph the pitch range and average pitch independently.

  • resynthesis is performed by Praat.

  • pitch can be obtained from praat (such as by using praatio) or from other sources (e.g. ESPS getF0)

  • plots of the resynthesis (such as the ones below) can be generated

2   Illustrative example

Consider the phrase "Mary rolled the barrel". In the first recording (examples/mary1.wav), "Mary rolled the barrel" was said in response to a question such as "Did John roll the barrel?". On the other hand, in the second recording (examples/mary2.wav) the utterance was said in response to a question such as "What happened yesterday".

"Mary" in "mary1.wav" is produced with more emphasis than in "mary2.wav". It is longer and carries a more drammatic pitch excursion. Using ProMo, we can make mary1.wav spoken similar to mary2.wav, even though they were spoken in a different way and by different speakers.

Duration and pitch carry meaning. Change these, and you can change the meaning being conveyed.

Note that modifying pitch and duration too much can introduce artifacts. Such artifacts can be heard even in pitch morphing mary1.wav to mary2.wav.

Pitch morphing (examples/pitch_morph_example.py):

The following image shows morphing of pitch from mary1.wav to mary2.wav on a word-by-word level in increments of 33% (33%, 66%, 100%). Note that the morph adjusts the temporal dimension of the target signal to fit the duration of the source signal (the source and generated contours are equally shorter than the target contour). This occurs at the level of the file unless the user specifies an equal number of segments to align in time (e.g. using word-level transcriptions, as done here, or phone-level transcriptions, etc.)

examples/files/mary1_mary2_f0_morph.png

With the ability to morph pitch range and average pitch, it becomes easier to morph contours produced by different speakers:

The following image shows four different pitch manipulations. On the upper left is the raw morph. Notice that final output (black line) is very close to the target. Differences stem from duration differences.

However, the average pitch and pitch range are qualities of speech that can signify differences in gender in addition to other aspects of speaker identity. By resetting the average pitch and pitch range to that of the source, it is possible to morph the contour while maintaining aspects of the source speaker's identity.

The image in the upper right contains a morph followed by a reset of the average pitch to the source speaker's average pitch. In the bottom right a morph followed by a reset of the speaker's pitch range. In the bottom right pitch range was reset and then the speaker's average pitch was reset.

The longer the speech sample, the more representative the pitch range and mean pitch will be of the speaker. In this example both are skewed higher by the pitch accent on the first word.

Here the average pitch of the source (a female speaker) is much higher than the target (a male speaker) and the resulting morph sounds like it comes from a different speaker than the source or target speakers. The three recordings that involve resetting pitch range and/or average pitch sound much more natural.

examples/files/mary1_mary2_f0_morph_compare.png

Duration morphing (examples/duration_manipulation_example.py):

The following image shows morphing of duration from mary1.wav to mary2.wav on a word-by-word basis in increments of 33% (33%, 66%, 100%). This process can operate over an entire file or, similar to pitch morphing, with annotated segments, as done in this example.

examples/files/mary1_mary2_dur_morph.png

3   Tutorials

Tutorials for learning about prosody manipulation and how to use ProMo are available.

Tutorial 1.1: Intro to ProMo

Tutorial 1.2: Pitch manipulation tutorial

4   Major revisions

Ver 1.3 (May 29, 2017)

  • added tutorials
  • f0Morph() can now exclude certain regions from the morph process if desired

Ver 1.2 (January 27, 2017)

  • added code for reshaping pitch accents (shift alignment, add plateau, or change height)

Ver 1.1 (February 22, 2016)

  • f0 morph code for modifying speaker pitch range and average pitch
  • (October 20, 2016) Added integration tests with travis CI and coveralls support.

Ver 1.0 (January 19, 2016)

  • first public release.

Beta (July 1, 2013)

  • first version which was utilized in my dissertation work

5   Requirements

Python 2.7.* or above

Python 3.3.* or above (or below, probably)

My praatIO library is used extensively and can be downloaded here

Matplotlib is needed if you want to plot graphs. Matplotlib website

Scipy is needed if you want to use interpolation--typically if you have stylized pitch contours (in praat PitchTier format, for example) that you want to use in your morphing). Scipy website

Matplotlib and SciPy are non-trivial to install, as they depends on several large packages. You can visit their websites for more information. I recommend the following instructions to install matplotlib which uses python wheels. These will install all required libraries in one fell swoop.

On Mac, open a terminal and type:

python -m pip install matplotlib

python -m pip install scipy

On Windows, open a cmd or powershell window and type:

<<path to python>> -m pip install matplotlib

<<path to python>> -m pip install scipy

e.g. C:\python27\python.exe -m install matplotlib

Otherwise, to manually install, after downloading the source from github, from a command-line shell, navigate to the directory containing setup.py and type:

python setup.py install

If python is not in your path, you'll need to enter the full path e.g.:

C:\Python27\python.exe setup.py install

6   Usage

See /examples for example usages

7   Installation

If you on Windows, you can use the installer found here (check that it is up to date though) Windows installer

Promo is on pypi and can be installed or upgraded from the command-line shell with pip like so:

python -m pip install promo --upgrade

Otherwise, to manually install, after downloading the source from github, from a command-line shell, navigate to the directory containing setup.py and type:

python setup.py install

If python is not in your path, you'll need to enter the full path e.g.:

C:\Python36\python.exe setup.py install

8   Citing ProMo

If you use ProMo in your research, please cite it like so:

Tim Mahrt. ProMo: The Prosody-Morphing Library. https://github.com/timmahrt/ProMo, 2016.

9   Acknowledgements

Development of ProMo was possible thanks to NSF grant BCS 12-51343 to Jennifer Cole, José I. Hualde, and Caroline Smith and to the A*MIDEX project (n° ANR-11-IDEX-0001-02) to James Sneed German funded by the Investissements d'Avenir French Government program, managed by the French National Research Agency (ANR).

Owner
Tim
I write tools for working with speech data.
Tim
Codes for “A Deeply Supervised Attention Metric-Based Network and an Open Aerial Image Dataset for Remote Sensing Change Detection”

DSAMNet The pytorch implementation for "A Deeply-supervised Attention Metric-based Network and an Open Aerial Image Dataset for Remote Sensing Change

Mengxi Liu 41 Dec 14, 2022
The implementation of the paper "HIST: A Graph-based Framework for Stock Trend Forecasting via Mining Concept-Oriented Shared Information".

The HIST framework for stock trend forecasting The implementation of the paper "HIST: A Graph-based Framework for Stock Trend Forecasting via Mining C

Wentao Xu 110 Dec 27, 2022
Meta-TTS: Meta-Learning for Few-shot SpeakerAdaptive Text-to-Speech

Meta-TTS: Meta-Learning for Few-shot SpeakerAdaptive Text-to-Speech This repository is the official implementation of "Meta-TTS: Meta-Learning for Few

Sung-Feng Huang 128 Dec 25, 2022
An open source object detection toolbox based on PyTorch

MMDetection is an open source object detection toolbox based on PyTorch. It is a part of the OpenMMLab project.

Bo Chen 24 Dec 28, 2022
Transfer Learning Remote Sensing

Transfer_Learning_Remote_Sensing Simulation R codes for data generation and visualizations are in the folder simulation. Experiment: California Housin

2 Jun 21, 2022
Saliency - Framework-agnostic implementation for state-of-the-art saliency methods (XRAI, BlurIG, SmoothGrad, and more).

Saliency Methods 🔴 Now framework-agnostic! (Example core notebook) 🔴 🔗 For further explanation of the methods and more examples of the resulting ma

PAIR code 849 Dec 27, 2022
It is an open dataset for object detection in remote sensing images.

RSOD-Dataset It is an open dataset for object detection in remote sensing images. The dataset includes aircraft, oiltank, playground and overpass. The

136 Dec 08, 2022
Few-Shot Graph Learning for Molecular Property Prediction

Few-shot Graph Learning for Molecular Property Prediction Introduction This is the source code and dataset for the following paper: Few-shot Graph Lea

Zhichun Guo 94 Dec 12, 2022
Vision Deep-Learning using Tensorflow, Keras.

Welcome! I am a computer vision deep learning developer working in Korea. This is my blog, and you can see everything I've studied here. https://www.n

kimminjun 6 Dec 14, 2022
Spatial Contrastive Learning for Few-Shot Classification (SCL)

This repo contains the official implementation of Spatial Contrastive Learning for Few-Shot Classification (SCL), which presents of a novel contrastive learning method applied to few-shot image class

Yassine 34 Dec 25, 2022
A PyTorch implementation of Sharpness-Aware Minimization for Efficiently Improving Generalization

sam.pytorch A PyTorch implementation of Sharpness-Aware Minimization for Efficiently Improving Generalization ( Foret+2020) Paper, Official implementa

Ryuichiro Hataya 102 Dec 28, 2022
NeWT: Natural World Tasks

NeWT: Natural World Tasks This repository contains resources for working with the NeWT dataset. ❗ At this time the binary tasks are not publicly avail

Visipedia 26 Oct 18, 2022
Code accompanying the paper on "An Empirical Investigation of Domain Generalization with Empirical Risk Minimizers" published at NeurIPS, 2021

Code for "An Empirical Investigation of Domian Generalization with Empirical Risk Minimizers" (NeurIPS 2021) Motivation and Introduction Domain Genera

Meta Research 15 Dec 27, 2022
Code for the TIP 2021 Paper "Salient Object Detection with Purificatory Mechanism and Structural Similarity Loss"

PurNet Project for the TIP 2021 Paper "Salient Object Detection with Purificatory Mechanism and Structural Similarity Loss" Abstract Image-based salie

Jinming Su 4 Aug 25, 2022
The PyTorch re-implement of a 3D CNN Tracker to extract coronary artery centerlines with state-of-the-art (SOTA) performance. (paper: 'Coronary artery centerline extraction in cardiac CT angiography using a CNN-based orientation classifier')

The PyTorch re-implement of a 3D CNN Tracker to extract coronary artery centerlines with state-of-the-art (SOTA) performance. (paper: 'Coronary artery centerline extraction in cardiac CT angiography

James 135 Dec 23, 2022
Physics-Aware Training (PAT) is a method to train real physical systems with backpropagation.

Physics-Aware Training (PAT) is a method to train real physical systems with backpropagation. It was introduced in Wright, Logan G. & Onodera, Tatsuhiro et al. (2021)1 to train Physical Neural Networ

McMahon Lab 230 Jan 05, 2023
Cluttered MNIST Dataset

Cluttered MNIST Dataset A setup script will download MNIST and produce mnist/*.t7 files: luajit download_mnist.lua Example usage: local mnist_clutter

DeepMind 50 Jul 12, 2022
Implementation of the CVPR 2021 paper "Online Multiple Object Tracking with Cross-Task Synergy"

Online Multiple Object Tracking with Cross-Task Synergy This repository is the implementation of the CVPR 2021 paper "Online Multiple Object Tracking

54 Oct 15, 2022
Implementation for NeurIPS 2021 Submission: SparseFed

READ THIS FIRST This repo is an anonymized version of an existing repository of GitHub, for the AIStats 2021 submission: SparseFed: Mitigating Model P

2 Jun 15, 2022
A Simulation Environment to train Robots in Large Realistic Interactive Scenes

iGibson: A Simulation Environment to train Robots in Large Realistic Interactive Scenes iGibson is a simulation environment providing fast visual rend

Stanford Vision and Learning Lab 493 Jan 04, 2023