What is Text-To-Speech?
Text-to-speech (or TTS, for short) is a type of speech synthesis, which is an artificially constructed voice created from snippets of human voice segments stringed together to create infinite number of sentences. It is called "text-to-speech" because the program can then convert what is typed into a computer into spoken language. Speech synthesis itself is the process of creating this functionality. It is what programs like Google Voice, Siri or Cortana or Echo use to communicate with the users.
What is TTS Yorùbá?
TTS in Yorùbá is a program to be created for the purpose of turning written Yorùbá text into spoken word/utterances.
Why Hasn't This Been Done Before?
There have been lots of research and activity on Speech Synthesis for Yorùbá but not many have been successful in being both intelligible and natural: two components that a good synthetic voice must have to be usable. According to Dr. Túndé Adégbọla, the director of the African Language Technology Initiative (ALT-I), in collaboration with Dr. Odetuji Odejobi in OAU Ife, they have "achieved appreciable levels of intelligibility of synthesized Yorùbá and we are still making good progress with naturalness." This is good news.
But the big commercial corporations with means to fund research into synthesis have only done it for the large 'marketable' languages, so Yorùbá and other "smaller" languages have been left behind.
How will it be done?
Our purpose for this project is to synthesize Yorùbá for words alone. And so, we would need a voiceover talent (VT) and a software development team.
This is what Dr. Adégbọlá of ALT-I has to say about our work: "With more than 5000 names already available on-line at the Yoruba names dictionary project web site, the idea of a narrow domain Yoruba names speech synthesizer is a most welcome next step, which will increase the functionality of the dictionary in quanta. It will turn the dictionary into a must-consult material for people all over the world who may need to find out the proper pronunciation of Yoruba names. I see it becoming a broadcasters' delight!"
Dr. Túndé Adégbọlá is the Director of the African Language Technology Initiative at ALT-I.org
Can YOU do it?
Yes, we can. When we launched the project in 2015, one of the element we spent time working on was this audio element. We have figured out the path to success, but we got stuck due to lack of funding and manpower. You can read more about that process in this blog post from 2015. You can also try out the result of that elementary attempt by clicking on these examples: Débọ́lá, Túndé, Folú, etc. You can also change the name on the web address to try out your own examples. They are not perfect yet but we know that we can get it right this time.
Why is this project important?
First, we need to create this software so as to convert all the names in the YorubaName.com Dictionary into audio. This way, users who do not speak Yorùbá but are interested in learning how to pronounce the names will be able to just click on the audio symbols and hear the pronunciation.
The only other way to convert all the 5000+ names in the dictionary into audio format is to have a human being pronounce them all. This is impractical for reasons of cost, time, and most importantly because of the dictionary's open-ended nature, which ensures that there is no upper limit to the number of names it will contain.
But we are considering research into many more ways in which this software can function in other domains in the future. Here are a few:
-
Accessibility applications
Users who can understand spoken Yorùbá but can't read it - users who are not literate in Yorùbá, blind and visually impaired people, people with other disabilities and young children - will be able to merely input the text into the machine and have it read out to them. TTS can be helpful to open up access to novels (audio books), scripture, text messages, and other texts for this demographic.
-
ATM services
ATM machines around Nigeria today only offer an English interface. A program like this can be incorporated in the software of ATM machines' artificial intelligence to serve customers in their mother tongue. If it can be done for Yorùbá, it can be done for other languages too.
-
Educational applications
As the synthesizer will be programmed to work only with the correctly-placed tone marks, this TTS software can be used in educational contexts to teach both orthography and pronunciation. This can have a net positive effect on the sustenance of the language.
-
Next step: automatic speech recognition
The converse side of this project is something called ASR (automatic speech recognition). If we can successfully train the computer to pronounce Yorùbá words properly from reading a text, then we might also be able to train it to understand Yorùbá speech and convert it to text. This will help in the areas of text dictation, and other AI situations like accessing phone or computer functionalities in Yorùbá (and other languages). See this in action at Mark Zuckerberg's house.
-
Expansion of multilingual tech applications
Availability of an innovative project usually spurs many other related ideas that can help create more innovations. The Nigerian space is in need of new ideas in language technology, and this project could be a trendsetter.
How Much Do We Need?
We are raising $4000 for this purpose, broken down as follows:
- $1500 for recording sessions (to include payment of voice-over talent, studio direction and post production)
- $1500 for software development to turn audio data into TTS program.
- $1000 any other unforeseen costs.
As usual, if we do not raise the complete amount, we will make do with whatever we get.
If you can help us in other ways, including but not limited to volunteering (especially with skills in software programming and architecture), please reach out at project@yorubaname.com with "TTS-Yorùbá" in the subject.
Can we donate in Naira or Paypal?
Yes, you may. Please send an email to project@yorubaname.com and we'll send you the appropriate bank account number for naira, and email address for PayPal contributions.
YorubaName.com is a research and documentation project which you can read more about here. The current effort is led by Kọ́lá, with support from Laila, and Victor. Meet the rest of the YorubaName team here.
