Transcript Model Provider is null

[Madaconda]

When I go to the Podcasts page, under Episodes or Templates, I see the following error:

KeyError: None
Traceback:

File "/app/.venv/lib/python3.12/site-packages/streamlit/runtime/scriptrunner/exec_code.py", line 121, in exec_func_with_error_handling
    result = func()
             ^^^^^^
File "/app/.venv/lib/python3.12/site-packages/streamlit/runtime/scriptrunner/script_runner.py", line 643, in code_to_exec
    _mpa_v1(self._main_script_path)
File "/app/.venv/lib/python3.12/site-packages/streamlit/runtime/scriptrunner/script_runner.py", line 164, in _mpa_v1
    page.run()
File "/app/.venv/lib/python3.12/site-packages/streamlit/navigation/page.py", line 297, in run
    exec(code, module.__dict__)
File "/app/pages/5_🎙️_Podcasts.py", line 118, in <module>
    transcript_provider_models[pd_cfg["transcript_model_provider"]],
    ~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

It might not be related but I have set TTS to be openai and have tried tts-1, tts-1-hd and gpt-4o-mini-tts. I have also set my openai api key in the .env and docker.env. I am running it in docker. Ideally I would like to be able to set it to use a local TTS or STT but it doesn't appear to be possible at this time?

Thanks for your time!

[lfnovo]

Hey @Madaconda.. there is no way to use local providers yet, although I have been working on some options for this to release soon.

While I work on a quick patch for your issue, can you please try deleting your podcast config and re-creating it?

[Madaconda]

In the process of trying to figure out how to delete my podcast config, I realized I wasn't able to even create one because of that same error. Then I noticed when I added openai gpt-4o to the language models, the error disappeared and I was able to see the templates / create templates. I had only set up local language models with ollama and I assume that doesn't work for a podcast transcript model? I wasn't able to test create a podcast because it said my openai api limit was reached. If there is a way of allowing this to work with ollama in the future and likewise with a local TTS, I would be grateful.
Thanks again for your time and work!

[lfnovo]

Hey Madaconda.. yes, the current podcast implementation does not support ollama for transcript generation but I am working on a work-around for it too. I'll leave a couple of issues for you to track this easily:

• Enable the use of all text models for transcript generation #84
• Enable local TTS and STT models #85