OpenRouter added separate interfaces for generating speech and transcribing recordings. Its audio API announcement distinguishes three jobs: a speech model turns supplied text into sound, a transcription model turns recorded speech into text, and a general audio model can reason over audio in a conversation. That separation lets an application request a specific conversion, such as reading text aloud or producing subtitles, while retaining OpenRouter's shared billing and access-key management. Developers choose the endpoint according to the required input and output, rather than treating every audio task as a conversation.
