Scripts for alignment of laboratory speech production data
- Kyle Gorman gormanky@ohsu.edu
- Michael Wagner chael@mcgill.ca
- FQRSC Nouvelle Chercheur NP-132516
- SSHRC Canada Research Chair 218503
- SSHRC Digging Into Data Challenge Grant 869-2009-0004
The MIT License
Copyright (c) 2011-2016 Kyle Gorman and Michael Wagner
Permission is hereby granted, free of charge, to any person
obtaining a copy of this software and associated documentation
files (the “Software”), to deal in the Software without
restriction, including without limitation the rights to use,
copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the
Software is furnished to do so, subject to the following
conditions:
The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
OTHER DEALINGS IN THE SOFTWARE.
Please you use this tool; we would appreciate if you cited the following paper:
Gorman, Kyle, Jonathan Howell and Michael Wagner. 2011. Prosodylab-Aligner: A Tool for Forced Alignment of Laboratory Speech. Canadian Acoustics. 39.3. 192–193.
USAGE: python3 -m aligner [OPTIONS]
Option Function
-c config_file Specify a configuration file to use [default: en.yaml]
-d dictionary Specify a dictionary file
(NB: available only with -t (See Input Group))
-h Display this message
-s samplerate (Hz) Samplerate for models [default: SAMPLERATE]
(NB: available only with -t)
-e Number of epochs in training per round [default: EPOCHS]
(NB: available only with -t (See Input Group))
-v Verbose output
-V More verbose output
Input Group: Only one of the following arguments may be selected
-r Read in serialized acoustic model
-t training_data/ Perform model training
Output Group: Only one of the following arguments may be selected
-a Directory of data to be aligned
-w Location to write serialized model
Forced alignment can be thought of as the process of finding the times at which individual sounds and words appear in an audio recording under the constraint that words in the recording follow the same order as they appear in the transcript. This is accomplished much in the same way as traditional speech recognition, but the problem is somewhat easier given the constraints on the "language model" imposed by the transcript.
The primary use of forced alignment is to eliminate the need for human annotation of time-boundaries for acoustic events of interest. Perhaps you are interested in sound change: forced alignment can be used to locate individual vowels in a sociolinguistic interview for formant measurement. Perhaps you are interested in laboratory speech production: forced alignment can be used to locate the target word for pitch measurement.
Yes! If you have a few hours of high quality speech and associated word-level transcripts, Prosodylab-Aligner can induce a new acoustic model, then compute the best alignments for said data according to the acoustic model.
Forced alignment works well for audio from speakers of similar dialects with little background noise. Aligning data with considerable dialect variation, or to speech embedded in noise or music, is currently state of the art.
You can train your own acoustic models, using as much training data as possible, or try to reduce the noise in your test data before aligning.
The Hidden Markov Model Toolkit (HTK) is a set of programs for speech recognition and forced alignment. The HTK book describes how to train acoustic models and perform forced alignment. However, the procedure is rather complex and the error messages are cryptic. Prosodylab-Aligner essentially automates the HTK forced alignment workflow.
The Penn Forced Aligner (P2FA) provides forced alignment for American English using an acoustic model derived from audio of US Supreme Court oral arguments. Prosodylab-Aligner has a number of additional capabilities, most importantly acoustic model training, and it is possible in theory to use Prosodylab-Aligner to simulate P2FA.
NB: when you are instructed to type in a command, do not type the '$' symbol; it just indicates the start of the prompt.
NB: most of these commands will produce significant text output. You can safely ignore it unless it explicitly is marked as an 'error'.
XCode is a free application that contains of all the tools you need to compile most software on Mac OS X. You can get it from the Mac App Store) or you can start the download from the Terminal. To do the latter, launch the application 'Terminal.app', then type the following at the prompt, then hit return:
$ xcode-select --install
Note that this is a large download and will take a while. Start it now! An alternative option that is somewhat smaller is Command Line Tools for Xcode.
Install HTK (Hidden Markov ToolKit)
HTK is the "backend" that powers the aligner. It is available only as uncompiled code. First, go to the HTK website and register. Then click on 'Download' on the left panel, and then click on 'HTK source code (tar+gzip archive)' under 'Linux/Unix downloads'.
Once this is downloaded, you may have to unpack the "tarball". Launch the application 'Terminal.app' (if you haven't already), and then navigate to your downloads directory (cd ~/Downloads will probably work). Then unpack the tarball like so:
$ tar -xvzf HTK-3.4.1.tar.gz
Some browsers automatically unpack compressed files that they download. If you get an error when you execute the above command, try the following instead:
$ tar -xvf HTK-3.4.1.tar
Once you extract the application, navigate into the resulting directory:
$ cd htk
Once this is complete, the next step is to compile HTK. Execute the following commands inside the htk directory:
$ export CPPFLAGS=-UPHNALG
$ ./configure --disable-hlmtools --disable-hslab
$ make clean # necessary if you're not starting from scratch
$ make -j4 all
$ sudo make -j4 install
(This will take a few minutes.)
At the last step, you may be asked to provide your system password; do so and then hit return. Note that your password will not echo (i.e., no '*' will be produced when you type).
Homebrew is a command-line application for installing software on your Macintosh. It is the easiest way to get the remaining dependencies to run the aligner. To install Homebrew, launch the application 'Terminal.app' (if you haven't already) and type the following:
$ ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
and then follow along with the instructions that are displayed in the terminal window. Once again, you may need to enter your system password, and once again, your password will not echo.
Homebrew makes it easy to install the newest version of Python programming language that powers the aligner. To install it, launch the application 'Terminal.app' (if you haven't already) and type the following:
$ brew install python3
(This will take a few minutes.)
SoX is the "Swiss Army knife of sound processing programs", and can be used to do fast batch of your audio files (though it is possible to run the aligner without using SoX). Once again, Homebrew makes it easy to install SoX. Launch the application 'Terminal.app' (if you haven't already) and type the following:
$ brew install sox
(This may take a few minutes.)
Prosodylab-Aligner lives on GitHub, a repository for open-source software. You may want to create an account there, and perhaps install 'GitHub.app', which makes it easier to interact with GitHub. But for the purposes of installing the aligner, all you need is the git command-line tool, which is part of Xcode (and so should already be installed). Launch the application 'Terminal.app' (if you haven't already) and type the following:
$ git clone http://github.com/prosodylab/Prosodylab-Aligner
Finally, you need to install a few additional dependencies for Python. Enter the following commands to take care of this:
$ cd Prosodylab-Aligner
$ pip3 install -r requirements.txt
At this point, you can test your installation by running:
$ python3 -m aligner --help
which should print out some information about how to use the aligner.
The instructions are the same as for Mac users, with the exception that there is no direct analogue to XCode. Instead, you will probably need to use your distribution's package manager to install a C compiler; on Ubuntu, for instance, the relevant package is called libc-dev.
While Prosodylab-Aligner is not designed for Windows support, the appendix of Yun et al. 2016 contains detailed Windows installation instructions. Note that while you may find these useful, they are are third-party instructions and we make no promises to their correctness.
The aligner comes with an (American) English dictionary file eng.dict. Some additional dictionaries we have created are available at prosodylab.dictionaries repository. Other dictionaries can be found online, or written for specific tasks. If you're working with RP speakers, CELEX might be a good choice. For languages with highly regular, transparent orthographies (e.g., Spanish or Tagalog), you may want to create a simple rule-based grapheme-to-phoneme system using a cascade of ordered rules.
Imagine you simply want to align multiple audio files with their associated label files, in the following format:
file data/myexp_1_1_1.*
data/myexp_1_1_1.lab: ASCII text
data/myexp_1_1_1.wav: RIFF (little-endian) data, WAVE audio, Microsoft PCM, 16 bit, mono 22050 Hz
cat data/myexp_1_1_1.lab
BARACK OBAMA WAS TALKING ABOUT HOW THERE'S A MISUNDERSTANDING THAT ONE MINORITY GROUP CAN'T GET ALONG WITH ANOTHER SUCH AS AFRICAN AMERICANS AND LATINOS AND HE'S SAID THAT HE HIMSELF HAS SEEN IT HAPPEN THAT THEY CAN AND HE'S BEEN INVOLVED WITH GROUPS OF OTHER MINORITIES
If you'd like to align multiple .wav/.lab file pairs, and they're all in a single directory data/, aligning them is as simple as:
$ python3 -m aligner -r lang-mod.zip -a data/ -d lang.dict
...
This will compute the best alignments, and then place the Praat TextGrids in the data/ directory.
The -r flag indicates the source of the acoustic model and settings to be used. In the example, lang-mod.zip represents the zip directory containing the acoustic model to be used.
-a data/ indicates the directory containing the data to be aligned.
-d lang.dict points to the dictionary to be used in aligning the data.
Secondly, a word in your .lab files may be missing from the dictionary. Such words are written to OOV.txt. You can transcribe these using a text editor, then mix them back in like so:
$ ./sort.py lang.dict OOV.txt > tmp;
$ mv tmp lang.dict
If you are transcribing new words using the CMU phone set, see this page for IPA equivalents.
Sometimes there are processing errors that occur. These can often be fixed by entering the following into Terminal:
$ make clean
$ export CPPFLAGS=-UPHNALG
$ ./configure --disable-hlmtools --disable-hslab
$ make -j4
$ sudo make -j4 install
Provide your password, if necessary.
The aligner module also allows you to train your own models,
$ python3 -m aligner -c lang.yaml -d lang.dict -e 10 -t lang/ -w lang-mod.zip
...
Please note: THIS REQUIRES A LOT OF DATA to work well, and further takes a long time when there is a lot of data.
When the -v or -V flags are specified, output is verbose. -v indicates verbose output while -V indicates more verbose output.
The -c flag points to the configuration file to use. In the example above, this file is lang.yaml. This file contains information about the setting preferences and phone set and is used to save the state of the aligner.
The -d flag points to the dictionary containing the words to be aligned.
The -w flag indicates that the resulting acoustic model and settings will be written to a file of the name following. In the example, the acoustic model and settings will be written to lang-mod.zip.
The -e flag is used to specify the number of training iterations per "round": the aligner performs three rounds of training, each of which take approximately the same time, so the effect of increasing this value by one is approximately 3-fold.
Lastly, the -t flag indicates the source of the training data. In the example, this is a directory called lang/. When -t is specified, a few other command-line options become available. The -s flag specifies samplerate for the models used, both training and testing data will be resampled to this rate, if they do not match it. For instance, to use 44010 Hz models, you could say:
$ python3 -m aligner -c lang.yaml -d lang.dict -e 10 -t lang -w lang-mod.zip -s 44010
...
Resampling this way can take a long time, especially with large sets of data. It is therefore recommended that samplerate specifications are made using resample.sh. This requires installing SoX (see above installation instructions).
To be more efficient, it is recommended that resample.sh is used to resample data. To do this, enter the following into your Terminal while in the aligner directory:
$ ./resample.sh -s 16000 -r data/ -w newDirectory/
The -s flag specifies the desired sample rate (Hz). 16000 Hz is the default for the aligner, and therefore recommended as a sample rate. Alternatively, a different sample rate can be specified for resample.sh and aligner module.
The -r flag points to the directory containing the files to be resampled.
The -w flag indicates the name of a directory where the new, resampled files should be written.