Error when normalizing stereo sound in groq_client.py

[shravankshenoy]

I think there is a minor bug in the code snippet below

https://github.com/PeoplePlusAI/sunva/blob/a32a67704d3a4750fe790ce27736015249614baa/backend/core/stt/groq_client.py#L75C8-L87

For stereo sound, when converting to mono, the datatype of audio_data becomes float64 because for np.mean the default return type for integer inputs is float64. On passing this float variable to np.iinfo, we will get an error as the iinfo function works only for integer types

Potential solution

1. Specify the output datatype as np.int16 in np.mean
   audio_data.mean(axis=1, dtype=np.int16)
2. Use np.finfo for stereo sounds

I think the first approach is better, but would love to know others thoughts

[bsbarkur]

@shravankshenoy so have you tested and run this hypothesis? Happy to see a PR from you to check the experiment.