gpt-4o-mini-tts pricing calculation could be very inaccurate

[p0n1]

Besides, the default tts model for OpenAI is changed from tts-1 to gpt-4o-mini-tts because it seems cheaper. But according to my testing, since the new model uses a new billing mode based on audio tokens, it seems to be more expensive for Chinese. So, be careful when using it.

FYI, my input tokens count during the test is 119,512 but the audio output tokens count is 756,986. I don't know how the audio output token is counted.

Originally posted by @p0n1 in #126

Related code

epub_to_audiobook/audiobook_generator/tts_providers/openai_tts_provider.py

Line 32 in f7f6b11

return 0.003 # TODO: this could be very wrong for Chinese. Not sure how openai calculates the audio token count.