Suttacentral voice to generate audio files from CSV?

I am toying with a project of making an audio anki deck of frequently used Pali words and later turning into an actual thing… like duolingo… maybe.
Is it possible to use SC-Voice for this sort of thing? Is there some documentation somewhere?

I’m starting with a list of 200 words. The words are all frequently used in the suttas so I assume they are quite well pronounced as opposed to my Australian sounding vowels.


No, very sorry, currently Voice is not able to read out single words out of context. You would have to go to AWS Polly. But there, pronunciation is not optimized as we have done it for Voice.

We had considered to add a “read custom Dhamma text” function, but then had too many other priorities … :person_shrugging:

All Voice documentation can be accessed via the “i” button in the top right corner. Here’s the English site. However, no guarantee that everything is up to date.


Thanks for your response.

What are the smallest chunks available? Is it just the same as the written segments?

1 Like

If one wanted to run Pali through AWS oneself, they could manually apply the IPA transformations from SC-Voice, no? I believe they live here and here?


The problem is that aws want a credit card so that I can use their tools.
I’ve been playing around with in the meantime but not getting very far.


Yeah! That’s using AWS Voices. So just select Raveena and give her some IPA: s̪ut̪t̪ə :slight_smile:
You have to use Raveena as she’s the only voice who can pronounce all the Pāli phonemes.

Is there a SC-Voice debugging tool where one can convert Pāḷi to IPA? That might be a nice tool to have! :slight_smile:

Edit: The Pāḷi script converter can convert Pāḷi to IPA. Testing now…

1 Like

Okay Using the Pāḷi Script converter to generate IPA, then plugging that into the ipa-reader app gives these results:

Visuddhimagga → ʋɪs̪ud̪d̪ʰɪməggə →

dhutaṅga → d̪ʰut̪əŋgə →

vipassanūpakkilesa → ʋɪpəs̪s̪ən̪uːpəkkɪleːs̪ə →

brahma-vihāra → bɾəɦməʋɪɦɑːɾə →

Brāhmaṇa → bɾɑːɦməɳə →

Not perfect. Visuddhimagga became more like visuddhīmigga to my ear… but not bad! Especially as a starting point :slight_smile: You can look through SC-Voice’s string replacements above for ideas on how to improve Raveena’s understanding of certain syllables


Yes, indeed.

Yes, you could. The relevant list is sc-voice/sc-voice/words/voices.json. But it only contains changes from what the voices naturally do.

You can get the entire strings of segments by applying the method that is documented here in the context of pronunciation editing (if you want to go there, please send me a PM so that I can give you access to the Voice admin section).

Yes, that’s exactly the problem. You need a credit card already for just creating an account. Then you can use their services for free for a year, but after that you are charged. (For working on pronunciation edits the scope of my bills was only cents, but still …).

Oh! :open_mouth:

:laughing: Yes!

1 Like

A great starting point thanks @Khemarato.bhikkhu

I was manually making IPA. I didn’t know there was a function in the script converter. Then I had an idea to use the script converter to sinhala characters and put that through a TTS.

Someone also wanted tts for the purpose of anki cards and made It accepts 200 characters at a time and generated individual mp3s for each new line in the box. I am grading the results of the audio generated and then I will custom build IPA for the ones I don’t like. So far it doesn’t like single syllable sounds with short vowels or words starting with ‘i’






Not bad really! Just the volume is a bit low.

1 Like