Text-to-speech inference API doesn't respect accept headers

Inference Endpoints on the Hub
1

When using the inference API using a restful HTTP call or the python library example, the returned audio is encoded in flac even though the accept headers have been set to wav or other supported encoding return type.

Anyone else have success getting different encodings?

2

Inference API or Inference Endpoints?

3

Inference API.

In fact the demo specifically writes/saves the file to example.wav but it is still actually encoded as .flac lol

4

Well found a possible culprit:

5

Updated the api-inference-community repo with the required changes with this Pull Request: