Meeting – Speech in and Speech Out!

This afternoon we debated the issues arising with speech recognition research and text to speech (TTS).  Mashael had two very interesting papers that were showing that Sphinx4 is still the place to be when it comes to looking at Arabic speech recognition but the debate about recognition rates with or without diacritics prevails.  In one paper it appeared that rates were higher without diacritic marks.

We then moved over to listen to the impact of diacritic marks with TTS.  Edrees had a web page that showed us how the recordings he had made with his mobile phone of two synthesised voices were clearer without diacritic marks.

اَللُغَةُ اَلعَرَبِيَةِ لُغَةٍ مَجِيْدَةٌ يَتَحَدْثُ بِهَاَ اَلنْاَسُ فِي أَكْثَرِ مَنْ سِتَةٍ وَ عِشْرِيْنَ دْوُلَةٍ حَوُلَ اَلعَاَلَمِ.

اَللُغَةِ اَلعَرَبِيَةِ مُفْرَدَاتُهَاَ غَيْرِ مَحْدُوُدَةٌ وَ تَحْتَوُيِ عَلَىَ عَدْدٍ مِنْ عَلَامَاتِ اَلْتَشْكِيِلِ اَلَتيِ تُمَيُزِ كَلِمَاَتِهْاَ وَ تَجْعَلَهَاَ لُغَةٌ مُعَقْدَةٌ بَعْضَ اَلشْئِ.


listen to .amr file with diacritic marks

اللغة العربية لغة مجيدة يتحدث بها الناس في أكثر من ستة و عشرين دولة حول العالم.

اللغة العربية لغة مفرداتها غير محدودة و تحتوي على عدد من علامات التشكيل التي تميز كلماتها وتجعلها لغة معقدة بعض الشئ.

listen to .amr file without diacritic marks


Arabic blog explains some of the issues around pronunciation and diacritics in a recent posting called ‘Arabic Diacritics (Al-Tashkeel الـتـشـكـيـــل )‘.

Further comments

I have taken the liberty of including a comment that Mashael made about our meeting in this blog as she has supplied us with some very useful links.

mashael on said: Edit

Hello All :)

After our meeting today me, Mrs. EA, and Edrees

Here are some papers which we think will be of interest

1) Arabic Phonetic Web Sites Platform Using VoiceXML : (Includes implementation of Arabic ASR (using Sphinx) and TTS (Using MBROLA project)

2)Natural speaker-independent Arabic speech recognition system based on Hidden Markov Models using Sphinx tools (What draws attention of this paper is that the system gives higher accuracy when implemented without diacritics)

3) This a dictation ASR developed by CMU university called EvalDictator (It supports Arabic, but it suffers from some problems. we’ll check how much progress have been archived on the project)

1 thought on “Meeting – Speech in and Speech Out!

Comments are closed.