Using deepfilter in Python on streaming audio chunks

[neeagl]

Will be great if someone can share pointers to be able to use DF to process continuous audio chunks coming from the microphone (or other sources)

Appreciate the help!
Thanks

[Safari77]

https://gist.github.com/Safari77/a505d2e6474240554c1f0acb4379a31f
https://raw.githubusercontent.com/spatialaudio/python-sounddevice/refs/tags/0.5.1/examples/plot_input.py
python ./plot_input.py -d "DeepFilter Noise Cancelling Source Mono"

[neeagl]

https://gist.github.com/Safari77/a505d2e6474240554c1f0acb4379a31f https://raw.githubusercontent.com/spatialaudio/python-sounddevice/refs/tags/0.5.1/examples/plot_input.py python ./plot_input.py -d "DeepFilter Noise Cancelling Source Mono"

@Safari77 Thanks for your quick revert. I'm looking for pure Python using native DF instead of the need to set up Pipewire. We'd need to deploy it to our app servers to process realtime audio feeds.

[neeagl]

I have tried using a similar implementation as this one: https://github.com/Sharrnah/whispering/blob/main/Models/STS/DeepFilterNet.py but unable to use it correctly.

[Safari77]

go to the git repo dir DeepFilterNet/pyDF and

python setup.py install --user
pip install deepfilternet

then start coding

from df.enhance import enhance, init_df, load_audio, save_audio

model, df_state, _ = init_df()

[neeagl]

Thanks @Safari77. The shared snippet works as a batch job on the given audio file but what I'm looking here is the ability to process small chunks of audio in real-time (20ms each). Basically will be feeding in chunks from Twilio to remove the bg noise before further processing.

from df.enhance import enhance, init_df, load_audio, save_audio

model, df_state, _ = init_df()

[whatscoding]

@neeagl I have the same request. Have you been able to make it work?

[neeagl]

@whatscoding unfortunately no. Still looking for options