What languages does the new TTS API support? Is there a list of the supported languages somewhere? I need it to speak with the correct accent.
I couldnât find anything in docs or the announcement regarding languages supported on the TTS models, so I tested it with a sentence of Punjabi (my language) written in Gurmukhi script and it surprisingly works. It sounds better than if I were to ask a native English speaker to read transliteration of the sentence.
Iâd share audio but this platform doesnât support audio files.
The TTS API has no language limitation (i.e. it supports any text in any language). However, youâll probably get better performance in some languages than others (such as English), reflective of the training data.
Hi @atty-openai , how about dialects or accents that share the same language (orthography) ? E.g. is there a way to explicitly ask for Australian English TTS instead of American English, or Cantonese instead of Mandarin Chinese?
Unfortunately not at the moment, but thatâs useful feedback, thank you!
I have not found that your TTS can support cantonese, So i was wondering can u help our product realize this feature?
Is there a roadmap of upcoming voices and local dialects (like French Canadian)?
Unfortunately the French output really sounds like an anglophone speaking in French. Excellent prosody and intonations. But definitely something we can use in production without offending a large portion of our customers.
Here is a list of all 57 languages supported by OpenAI thanks to Microsoft Neural Speech Studio.
Afrikaans (South Africa)
Arabic (Egypt)
Armenian (Armenia)
Azerbaijani (Azerbaijan)
Belarusian (Belarus)
Bosnian (Bosnia and Herzegovina)
Bulgarian (Bulgaria)
Catalan (Spain)
Chinese (Mandarin, Simplified)
Croatian (Croatia)
Czech (Czech)
Danish (Denmark)
Dutch (Netherlands)
English (United States)
Estonian (Estonia)
Finnish (Finland)
French (France)
Galician (Galician)
German (Germany)
Greek (Greece)
Hebrew (Israel)
Hindi (India)
Hungarian (Hungary)
Icelandic (Iceland)
Indonesian (Indonesia)
Italian (Italy)
Japanese (Japan)
Kannada (India)
Kazakh (Kazakhstan)
Korean (Korea)
Latvian (Latvia)
Lithuanian (Lithuania)
Macedonian (North Macedonia)
Malay (Malaysia)
Marathi (India)
MÄori (New Zealand)
Nepali (Nepal)
Norwegian (BokmÄl, Norway)
Persian (Iran)
Polish (Poland)
Portuguese (Brazil)
Romanian (Romania)
Russian (Russia)
Serbian (Serbia)
Slovak (Slovakia)
Slovenian (Slovenia)
Spanish (Spain)
Swahili (Kenya)
Swedish (Sweden)
Filipino (Philippines)
Tamil (India)
Thai (Thailand)
Turkish (TĂŒrkiye)
Ukrainian (Ukraine)
Urdu (Pakistan)
Vietnamese (Vietnam)
Welsh (United Kingdom)
Can anyone share the complete list of Indian languages supported by the OpenAI Realtime API? I couldnât find an official list in the documentation.
gpt-realtime-2 is out as of the last two months (not the last two years).
OpenAI has testimonials there from an Indian company that got a preview before us,
In our evals across Hindi, Tamil, and Telugu, GPT-Realtime-Translate delivered 12.5% lower Word Error Rates
For a listing, this raw cookbook page has the same 70+13 claims as the gpt-realtime translate 2 version announcement, yet doesnât have Tamil:
https://raw.githubusercontent.com/openai/openai-cookbook/main/examples/voice_solutions/realtime_translation_guide.mdx
Language speaking is a somewhat emergent ability, such as the models being able to write some Cherokee or Inuktitut despite no specific large corpus. Voice is expected to be the same, based not merely on a listing, but a quality cutoff you need to determine yourself.