Building a phonetic dictionary



Introduction

A phonetic dictionary provides the system with a mapping of vocabulary words to sequences of phonemes. It might look like this:

hello H EH L OW
world W ER L D

A dictionary can also contain alternative pronunciations. In that case you can designate them with a number in parentheses:

the TH IH
the(2) TH AH

There are various phonesets to represent phones, such as IPA or SAMPA. CMUSphinx does not yet require you to use any well-known phoneset, moreover, it prefers to use letter-only phone names without special symbols. This requirement simplifies some processing algorithms, for example, you can create files with phone names as part of the filenames without any violating of the OS filename requirements.

A dictionary should contain all the words you are interested in, otherwise the recognizer will not be able to recognize them. However, it is not sufficient to have the words in the dictionary. The recognizer looks for a word in both the dictionary and the language model. Without the language model, a word will not be recognized, even if it is present in the dictionary.

There is no need to remove unused words from the dictionary unless you want to save memory, extra words in the dictionary do not affect accuracy.

Using existing dictionaries

There are a number of dictionaries which cover languages we support – CMUDict for US English, French, German, Russian, Dutch, Italian, Spanish and Mandarin. Other dictionaries might be found on the web. If a dictionary has a proper format you can use it.

If a dictionary does not cover all the words you are interested in, you can extend it with the g2p tool.

Using g2p-seq2seq to extend the dictionary

There are various tools to help you to extend an existing dictionary for new words or to build a new dictionary from scratch. Two of them are Phonetisaurus and Sequitur.

We recommend to use our latest tool g2p-seq2seq . It is based on neural networks implemented in the Tensorflow framework and provides a state-of-the-art accuracy of conversion.

An English model 2-layer LSTM with 512 hidden units is available for download on the CMUSphinx website. Unpack the model after downloading. It is trained on the CMU English dictionary. As the name says, this model works only for English. For other languages you first need to bootstrap a dictionary as described below and then use the G2P tool to extend it.

The easiest way to check how the G2P tool works is to run the interactive mode with the model from above:

g2p-seq2seq --interactive --model model_folder_path

> hello
HH EH L OW

To generate pronunciations for an English word list with a trained model, run:

g2p-seq2seq --decode your_wordlist --model model_folder_path

The wordlist is a text file with words, one word per line.

To train G2P you need a dictionary (a word and phone sequence per line in the standard form). Run the training with:

g2p-seq2seq --train train_dictionary.dic --model model_folder_path

For more information on the G2P tool have a look at the Readme of the G2P project.

Bootstrapping a dictionary for other languages

If you do not have a dictionary for your language there are usually several ways how you can create them.

Usually, dictionaries are bootstrapped with hand-written rules. You can find a list of phonemes for your language in the Wikipedia page about your language and write a simple Python script to map words to phonemes. The best dictionary could not be covered with rules though, most languages have quite irregular pronunciation which might not be very obvious for a newcomer even if it is conventionally thought that you speak what is written. This is due to coarticulation effects in human speech. However, for a basic dictionary, rules are sufficiently good enough.

You can crawl the Wiktionary to get a mapping for a significant amount of words covered there.

You can use TTS tools like from OpenMary written in Java or from Espeak written in C to create the phonetic dictionary for the languages they support.

Many languages which use hieroglyphs like Korean or Japanese have specialized software like Mecab to romanize their words. You can use Mecab to build a phonetic dictionary by converting words to the romanized form and then simply applying rules to turn them into phones.

It is enough to transcribe a few thousand most common words to bootstrap the dictionary.

Once your dictionary is bootstrapped you can extend it to hold a larger vocabulary with the g2p-seq2seq tool as described in the previous section.

Using PocketSphinx on Android Building a language model