Forum Discussion

MichelleDoyle-a's avatar
MichelleDoyle-a
Community Member
19 days ago

AI tex-to-speech

We are currently evaluating the trial version of the AI text-to-speech functionality. We are wondering if additional accents will be available in the non-trial version. We also need Vietnamese and Thai; what is the likelihood of those being supported in the future? We tried to use the Czech but seemed to be having issues. Is there a voice in the trial version that we should use to have the best experience? 

This is the current list that I found of supported languages:

Arabic (Saudi Arabia)
Arabic (UAE)
Bulgarian
Chinese
Croatian
Czech
Danish
Dutch
English (Australia)
English (Canada)
English (UK)

English (USA)

Filipino
Finnish
French (Canada)
French (France)
German
Greek
Hindi
Indonesian
Italian
Japanese
Korean
Malay
Polish
Portuguese (Brazil)
Portuguese (Portugal)
Romanian
Russian
Slovak
Spanish (Mexico)
Spanish (Spain)
Swedish
Tamil
Turkish
Ukrainian

 

With this list, I am not sure which voices to use as most of the voices available in the trial version seem to have an "American" accent. 

 

Please advise. 

Michelle Doyle

ai assistant
e-learning development
Rise 360
storyline 360
  • AnnieKim18's avatar
    AnnieKim18
    Staff
    18 days ago

    My colleague found that on their site, they list Alice for Czech so that's probably the best voice to use for now until they add more. 

  • AnnieKim18's avatar
    AnnieKim18
    Staff
    18 days ago

    Hi Michelle, we use ElevenLabs for our AI text-to-speech voices and are currently using the Multilingual v2 model which supports the languages you listed above. According to their documentation, these voices can be used to generate audio in any of the languages, even though they are listed with an American accent. It also states that they're working to provide a granular overview of how each default voice sounds in each of the languages and will update that page when it's ready. They currently don't support Thai, but they do support Vietnamese using a different model (Turbo v2.5) which we are not using because it values generation speed over accuracy. But I will note this request as we consider switching to the Turbo v2.5 model or supporting multiple models in the near future.

Related Content