نوع مقاله: مقاله پژوهشی

نویسندگان

1 گروه آموزشی زبان‌شناسی،دانشکده‌ی زبان‌های خارجی، دانشگاه اصفهان، اصفهان، ایران.

2 گروه آموزشی زبان شناسی، دانشکده زبان های خارجی، دانشگاه اصفهان، اصفهان، ایران

3 گروه هوش مصنوعی، دانشکده کامپیوتر، دانشگاه اصفهان، اصفهان، ایران

4 دانشکده‌ی ریاضی، آمار و علوم کامپیوتر، دانشگاه تهران، تهران، ایران.

چکیده

با توجه به نقش اساسی پیکره‌ها در بهبود کیفیت عملکرد سیستم‌های مبتنی بر داده‌، به‌کارگیری پیکره‌های گفتاری مناسب در سیستم‌های پردازش گفتار نیز امری ناگزیر است. به طور معمول در سیستم‌های پردازش گفتار، از پیکره‌هایی استفاده می‌شود که جملات آن در سطح کلمه و واج، تقطیع شده‌ است. یکی از روش‌های مطرح جهت افزایش دقت سیستم‌های پردازش گفتار در سال‌های اخیر، استفاده از پیکره‌های واجگونه‌ای در این سیستم‌هاست. همان‌گونه که از نام آن‌ها پیداست، ویژگی بارز پیکره‌های واجگونه‌ای نسبت به پیکره‌های واجی، اختصاص برچسب‌هایی ویژه (یعنی همان برچسب‌های واجگونه‌ای) به هر یک از واج‌های تقطیع ‌شده است. راهکار پیشنهادی برای تهیه‌ی پیکره‌ی واجگونه‌ای، پیاده‌سازی برنامه‌ای با استفاده از روش مبتنی بر قاعده، برای تبدیل واج‌ها به واجگونه‌ها و اعمال این برنامه بر پیکره‌ی واجی، به منظور برچسب‌گذاری واجگونه‌ای آن است. شالوده‌ی استفاده از رویکرد مبتنی بر قاعده در این پژوهش، دسترسی به قواعدی برای تبدیل واج‌ها به واجگونه‌هاست. پس از تدوین این قواعد از منابع موجود در هر زبان، ایجاد بستر مناسب (جهت پیاده‌سازی) و سپس پیاده‌سازی برنامه‌ی مربوطه و در نهایت، اعمال این برنامه بر پیکره‌ی گفتاری واجی، پیکره گفتاری واجگونه‌ای تهیه می‌شود. زبان فارسی نیز فاقد پیکره‌ی واجگونه‌ای است و پیکره‌ی گفتاری فارس‌دات کوچک در این زبان، دارای تقطیع در سطح واج و کلمه است. به منظور تبیین هر چه بهتر راهکار پیشنهاد شده در این پژوهش، به عنوان یک نمونه‌ی عملی، مراحل برچسب‌گذاری واجگونه‌ای پیکره‌ی واجی فارس‌دات کوچک، به صورت گام به گام اجرا شده است.

کلیدواژه‌ها

عنوان مقاله [English]

Providing a suitable method for allophonic labeling of speech corpuses according to the IPA system

نویسندگان [English]

  • Tahere Ahmadi 1
  • Batool Alinezhad 2
  • Hossein Karshenas 3
  • Bagher Babaali 4

1 linguistics Department, foreign languages Faculty, Isfahan University , Isfahan, Iran.

2 Linguistic Department, Foreign Languages Faculty, Isfahan University, Isfahan, Iran

3 Artificial intelligence Department, computer Faculty, Isfahan university, Isfahan, Iran

4 Faculty of Mathematics, Statistics and Computer Science, University of Tehran, Tehran, Iran.

چکیده [English]

Abstract
The corpus is a collection of spoken and / or written texts that can be used for linguistic analysis. More precisely, it can be said that these texts are purposefully labeled and categorized based on specific rules and allow the user to do various studies. Corpus linguistics is a branch of applied linguistics that examines and compares different aspects of linguistic data, and, of course, corpuses are an integral tool of this branch of linguistics. Due to the increasing role and importance of corpus linguistics in the development of various sciences in recent decades, the produce and development of various linguistic corpuses has been one of the priorities of scientists and researchers in different languages during these years.
After the creation of speech processing systems since about two decades ago, in recent years, and in an effort to increase the accuracy of these systems, and in addition to conducting some special studies in linguistics, the use of context-dependent methods has become particularly prominent. One of the best ways to achieve this, is to use corpuses that, in addition to segmentation at the phoneme level, have special labels to indicate the differentiation of various allophones (which can only be achieved by obtaining the necessary phonological rules). In linguistic, this process can be called allophonic labeling of corpus.
About 10 years after the introduction of allophonic corpuses in the world, no allophonic labeling has been performed for any of Persian language corpuses yet. The small Farsdat corpus is the main spoken corpus in Persian. So the need to equip it with allophonic labels to increase the accuracy and improve the performance of speech processing systems as well as the production of specific study and research programs and tools in linguistic, is obvious. In order to elucidate the method proposed in the present study for allophonic labeling of phonemic corpuses, and in parallel for equipping the Persian language with at least one allophonic corpus, the steps of the task are precisely performed on the small Farsdat phonemic corpus.
Small Farsdat corpus is one of Persian-language corpuses in the last two decades. This corpus consists of 6080 sentences spoken by 304 Persian speakers (that have one of the most widely spoken dialects in Persian) and all of sentences in this corpus, is segmented in to different levels. The segmentation of sentences in word and phoneme levels results in their efficiency in various speech processing systems such as speech recognition systems, broad transcription systems, and text-to-speech systems; and the small Farsdat corpus also has the potential to be used in the systems.
The suggested solution to prepare an allophonic corpus, is to implement a program using the rule-based method, and apply it on the phonemic corpus to add allophonic labels on it. The basis of the rule-based method in this research is access to rules for converting phonemes into allophones. After compiling these rules from the resources available in each language and preparing the appropriate settings (for implementation) implementing the program is done. Finally by applying this program to the phonemic corpus, an allophonic corpus is prepared.
As noted, special phonological rules are required to convert phonemes in to allophones in Persian and then add allophonic labels to the small Farsdat corpus. The purpose of this research, is not acoustic and laboratory study on phonemes in order to obtain Persian allophones; but rather to formulate and synchronize phonemes identified in various studies and then to adapt them to the International Phonetic Alphabet System. This ultimately leads to provide a standard set of allophones and, as far as possible achieve the phonological rules necessary for converting phonemes into allophones in Persian (based on existing studies so far).
Although one of the limitations of this study is the incomplete studies on the extraction of different allophones in Persian, the implemented program has the capability to be updated; and if carry out any studies in the field of allophones and supplement the existing theoretical resources in the future, it can be changed and Has the ability to modify or enhance performance based on new studies. The present study may also highlight the need for more recent linguistic experiments and the use of more accurate tools and facilities to identify Persian phonemes and increase the motivation of phonetics and phonology researchers to take more practical steps in this field.
After providing the necessary preparations in the phonemic corpus (such as the syllable segmentation) and implementing the above rules, the allophonic labels can be added to the phonemic corpus by implementing this program on it.

کلیدواژه‌ها [English]

  • phoneme
  • allophone
  • corpus
  • IPA system