Resources for Arabic, specifically in the Palestinian dialect are scarce. For a while now, I’ve thought about using TTS to generate Palestinian Arabic audio for learning purposes. The results though have not been very good.
For example, here is a simple sentence in Arabic, without tashkeel (the diacritics used to present vowels for pronunciation):
هادا بيت جديد و هادا كمان بيت جديد
The common problem I face with Arabic TTS is that it’s all formal MSA1. I then found out that Elevenlabs, a speech generation platform, supports users uploading their own voices and accents for others to use. Maybe there could be a Palestinian Arabic dialect voice out there for me to use?
I was able to find 2 voices that are specifically tagged as Palestinian Arabic:
The results though, were inconsistent, even with the very simple example sentence:
Example generated with Elevenlabs using the Multilingual v2 model with a Palestinian dialect voice.
In this example, the sentence is spoken in a formal, Modern Standard Arabic (MSA/Fusha) form, not the spoken form of Palestinian Arabic that I was looking for.
Recently however, Elevenlabs released Eleven v3. Even though it doesn’t seem to support custom voices (in this case the Palestinian dialect custom voice), the base model did a much better job for the same sentence:
Example generated with Eleven v3.Testing a few other sentences as well, I am quite impressed with the results. For now my own skills in Arabic are not sufficient to say exactly how accurate the speech is for more complex sentences. But there is potential here.
I am already planning on supplementing my Palestinian Arabic Anki decks with this audio to aid learning. Will share more on that later.
Footnotes
-
I guess it makes sense for them to be formal MSA, but it’s not useful for my learning purposes ↩