Hi everyone,
I want to report an issue with how Persian voice output is handled in both ChatGPT and Sora. It affects accuracy for most Persian speakers.
The issue
When making API requests for Persian voice output, especially spoken dialogue inside Sora-generated videos, the system consistently defaults to the Afghan Dari dialect. This occurs even when the request simply asks for “Persian” or “Farsi,” which in almost all global contexts refers to Iranian Persian.
To get the standard Iranian or Tehrani dialect, you must explicitly request it, which should not be necessary for API usage.
The voice output uses Dari-style pronunciation, cadence, and vocabulary that most Persian speakers do not typically use or immediately recognize.
Why this matters
Although Dari is a legitimate dialect of Persian, it is not the form spoken by most Persian speakers. The majority of Persian speakers worldwide use the Iranian or Tehrani dialect. Dari differs enough in sound and vocabulary that many Iranian speakers find it unfamiliar or difficult to follow. Having Dari as the default Persian voice is unusual and does not match common expectations.
This is especially noticeable in Sora videos, where natural and familiar speech is important.
Minimal reproducible examples (API context)
ChatGPT Voice
Input: "Please speak in Persian."
Output: Dari pronunciation and vocabulary
Expected: Standard Iranian (Tehrani) Persian voice
Sora
Input: "Generate a video with a speaker talking in Persian."
Output audio: Dari pronunciation and vocabulary
Expected: Standard Iranian (Tehrani) Persian voice
Possible reasons
This may be related to the Persian-language voice recordings that have been most widely available for training. A large amount of public material comes from Afghanistan, including aid-related projects and language resources created for people working there. If much of the training data labeled as Persian is actually Dari, the system may have ended up learning that as the default.
Request
Please set the default Persian (Farsi) voice to standard Iranian Persian, since this reflects the majority of actual Persian speakers and aligns with common user expectations (and needs). The Dari voice should, of course, remain available, but only when it is specifically requested.
Thanks for reviewing this.