Silent output #6
Description
I'm trying to experiment using query-bandit on some orchestral dat but unfortunately I'm not able to produce any output. Every output file is silent. I've tried the recommended ev-pre-aug.ckpt together with different query audios of about 30s.
This is my inference command, similar to the README version:
python banquet/train.py inference_byoq --ckpt_path="banquet/ckpt/ev-pre-aug.ckpt" --input_path="example/symphony.wav" --output_path="example/estimate/violin.wav" --query_path="example/Violin_1.wav" --config_path="banquet/expt/bandit-everything-test.yml" --batch_size=4 --use_cuda=trueI've analyzed the audio statistics and have seen that in the "chunked_inference" function the chunked_audio and the query audio have standard deviations of about 0.025-0.3 but the output signal has 0.00001. So there is really no signal coming out of the model.
I've tried different input audios and also different formats.
When I try ev-pre-fz-bal.ckpt the whole thing works.