ChatGPT produces amazing natural-sounding voice and supports multiple languages, I think it would significantly improve user experience. Hope developer team will consider it!
https://platform.openai.com/docs/guides/text-to-speech
9 Likes
Yes please!!! 
This is the #1 thing I wish was better on LingQ.
1 Like
Those voices are impressive! But, they aren’t free. I suspect it would be cost prohibitive for LingQ to include them unless they found a way to pass through the cost or link to a user’s paid account.
2 Likes
I would love for Lingq to integrate a new TTS system that is better than the current one, but if I’m not mistaken, all the demos of ChatGPT TTS API (including those from OpenAI, developers testing the API, and using TTS within ChatGPT app) that I have seen in different languages have a distinct English accent. I believe it is not mature enough to be used on a language learning platform like Lingq. .
2 Likes
@srdurden makes a good point. I only listened to the voices in English.
Regarding cost, just now I set up a small chatGPT account (having used up my free account already a long time ago). You can do this with as little as $5.
Using the API I generated an mp3 of the shortest chapter in the book I am reading, about 340 words and about 2 minutes long, in French. It cost me 4 cents. Which seems like nothing! but would quickly become costly for LingQ unless they passed the cost through somehow.
I’m going off to listen to it now…
2 Likes
@kindl How did the API work in French? Did it have an English accent? The ChatGPT app allows reading aloud texts (which could be saved in video to extract the audio) but I can’t find a way for it to read texts without an accent.
1 Like
@sdurden The accent sounded fine to me although there may be nuances I miss as an American English speaker. The API documentation says that multiple languages are supported although it states that things are optimized for English right now.
The TTS technology is apparently based on Whisper. Someone analyzed Whisper across many languages and found it is best at Spanish, Italian, Portuguese and worst at Swahili, Welsh, Belarusian (this is a gross generalization of mine from looking at the tables for two models in the Whisper GitHub. There’s a whole paper on the study). French is toward the top as is English so one’s mileage will likely vary depending on the language being read.
Here is how the Chat GPT API doc explains the cross language capability:
“ The TTS model generally follows the Whisper model in terms of language support. Whisper supports the following languagesand performs well despite the current voices being optimized for English:
Afrikaans, Arabic, Armenian, Azerbaijani, Belarusian, Bosnian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, German, Greek, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Kannada, Kazakh, Korean, Latvian, Lithuanian, Macedonian, Malay, Marathi, Maori, Nepali, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swahili, Swedish, Tagalog, Tamil, Thai, Turkish, Ukrainian, Urdu, Vietnamese, and Welsh.”
1 Like
@kindl Could you please upload the audio clip somewhere to see if French speakers here notice an accent?
1 Like
@srdurden I have posted the first part of Chapter 1 of The Wizard of Oz (Le Magicien d’Oz) in Google Drive here: wizard.mp3
The title of the chapter is “Le cyclone”. I see that I should have forced a pause before the first sentence, with a comma or period or something.
This book is in the public domain. The TOU of the API requires that I state that this narration was created by AI and is not a person.
To my ear the accent is good. It is certainly not an American accent.
1 Like