Starting around 24 hours ago, in about 1 of 20 requests, tts-1-hd randomly returns an empty response (no audio). Our system then retries after 1 minute with the same text and gets a valid audio response. Our logs are full of these exceptions now.
Hi,
I could reproduce this too. After around the 10th request, I got an empty response.
Since it seems intermittent, request IDs and timestamps would probably be important here so OpenAI staff can check the logs. It would also help to note whether retrying with the same text consistently returns valid audio.
In other words, I’d recommend opening a support ticket for this and including the request IDs/timestamps there as well.
Also there was another report yesterday regarding gpt-4o-mini-tss being slow. I don’t know if these issues could be related.
I do not know, if the first retry with the same text returns valid audio. But I know for shure, that at least after a few retries a valid audio was generated for every text (several hundred paragraphs). Our system was able to generate audio for all texts because it retries. None of the texts were skipped or modified.
Thanks for reporting this. The thread confirms that retries eventually generated audio, but it does not confirm that the underlying issue was fixed.
If you still see an empty or no-audio response from tts-1-hd, please provide one recent failed x-request-id with its exact timestamp and timezone. Please also confirm the response size and clarify whether the response was empty, rather than a valid audio file containing silence.
Do not share API keys, authorization headers, or other secrets. These details can help Support investigate the failed request and distinguish the original symptom from later silent-audio reports.