[InferenceClient - Automatic speech recognition] Infer content type header based on file extension

[Wauplin]

Follow-up issue after #2706. The initial intent of the issue is still left to be done. Sending a content-type as header can be done like this:

client = InferenceClient(url, headers={"Content-Type": "audio/mpeg"})
response = client.automatic_speech_recognition("audio.mp3")

would be even better to automatically infer the content type header based on file extension. Only possible when input is passed as a filepath, not raw bytes.

Note: this is not specific to automatic_speech_recognition but to any "binary-only" task (AST, audio-to-audio, image classification, image-to-xxx, etc.). Only useful when binary is sent alone, not when sent as base64-encoded with other parameters.

[WizKnight]

Hi @Wauplin🤗, I would love to contribute to this.
Looking forward to your response!

Best regards