Improvements

[balthazarneveu]

:gift: A few bonuses

• [x] Neutral network
• [ ] correct timeline in interactive pipe
• [ ] epoch slider - :gift: Nice to have... not really necessary
• [x] display output metrics live in the GUI (out SNR)
• [ ] infer any voice in the batch Note: :dart: you can infer a premixed version with scripts/audio_separation_batch.py but not remix two signal (like -s signal.wav -n noise.wav)

Architecture tests

• [x] try bias free architecture bias free

[balthazarneveu]

:-1: Bonuses - useless stuff

• Tensoboard :gift: bonus - W&B is pretty good already
• Dumps a few audio at test phase (during trainings)

[balthazarneveu]

https://github.com/balthazarneveu/gyraudio/assets/41070742/76803e54-c614-4a3c-9cae-5d967ecf87ad

https://github.com/balthazarneveu/gyraudio/assets/41070742/a6b02e90-979d-4b06-a5b9-d8590c8c7692

https://github.com/balthazarneveu/gyraudio/assets/41070742/d0f01bc7-4868-4cad-a961-eb73748ccdfa

https://github.com/balthazarneveu/gyraudio/assets/41070742/c8940622-d0fc-4e2c-9409-2d0e4b8285c6