Spleeter: [Question] Training error

[cheffplayer]

I looked at your wiki and saw that it was possible to train a new model with our own data. I have went about setting up everything that the wiki states, but the furthest I can get is this error message:

TypeError: x and y must have the same dtype, got tf.string != tf.int32

The command I am executing is

spleeter train -p musdb_config.json -d D:\spleeter

My config file is as follows:

{
"train_csv": "musdb_train.csv",
"validation_csv": "musdb_validation.csv",
"model_dir": "musdb_model",
"mix_name": "mix",
"instrument_list": ["synth", "drums", "guitar"],
"sample_rate":44100,
"frame_length":4096,
"frame_step":1024,
"T":512,
"F":1024,
"n_channels":2,
"n_chunks_per_song":1,
"separation_exponent":2,
"mask_extension":"zeros",
"learning_rate": 1e-4,
"batch_size":4,
"training_cache":"cache/training",
"validation_cache":"cache/validation",
"train_max_steps": 100000,
"throttle_secs":600,
"random_seed":3,
"save_checkpoints_steps":300,
"save_summary_steps":5,
"model":{
"type":"unet.unet",
"params":{
"conv_activation":"ELU",
"deconv_activation":"ELU"
}
}
}

Both my .csv are in correct format

drums_path,guitar_path,synth_path,mix_path,duration
train/christmas2drums.wav,train/christmas2guitar.wav,train/christmas2synth.wav,train/christmas2.wav,205.872‬

I have no idea what I am doing wrong, I wish there was a step-by-step tutorial on training our own models. Also, I am using the most latest version of spleeter-gpu installed by conda

[aidv]

You're not the only one having issues with training, I do too, and it seems like the Spleeter dev team has no interest in helping the community :/

[mmoussallam]

Hi @cheffplayer

It's not easy for us to help you if you don't provide the details we ask for in the Bug issue template, namely : what is the platform you use? how did you install spleeter ? what exact code are you running to get this message ?

[aidv]

@cheffplayer did you manage to solve this?