Training Speaker_ID SpeechBrain

Describe the bug

I am encountering difficulties while attempting to train the SpeechBrain speaker identification model using my own WAV files. I have made some modifications to the existing code provided in the train.yaml file, but I'm not sure if I'm even modifying the train.py and train.yaml files correctly. I have also saved a json file with annotations in the required format for training.

So far, the code runs well, but nothing is in the 'save' file. I'm hoping to use this code for speaker diarization. It would be great to know what I'm doing correctly and incorrectly to train the model.

Expected behaviour

I was expecting to have a trained model saved in the 'save' file after running %cd /content/speechbrain/templates/speaker_id !python train.py train.yaml --number_of_epochs=15 #--device='cpu' with modified train.py and train.yaml files. I was also expecting to use this model for speaker diarizaitno.

To Reproduce

So far, I've set the data_folder parameter in the train.yaml file to point to my own data directory. In the train.py file, I've updated the prepare_mini_librispeech function call in the main code section to my own data directory.

sb.utils.distributed.run_on_main(
    prepare_mini_librispeech,
    kwargs={
        "data_folder": "file_path",
        "save_json_train": hparams["train_annotation"],
        "save_json_valid": hparams["valid_annotation"],
        "save_json_test": hparams["test_annotation"],
        "split_ratio": hparams["split_ratio"],
    },
)

I've also modified the dataio_prep function to read my own WAV files from the specified directory and replaced the audio_pipeline function with the following code:

import glob

@sb.utils.data_pipeline.takes("file_path")
@sb.utils.data_pipeline.provides("sig")
def audio_pipeline(file_path):
    """Load the signal, and pass it and its length to the corruption class.
    This is done on the CPU in the `collate_fn`."""
    sig = sb.dataio.dataio.read_audio(file_path)
    return sig

# Define datasets. We also connect the dataset with the data processing
# functions defined above.
datasets = {}
data_info = {
    "train": hparams["train_annotation"],
    "valid": hparams["valid_annotation"],
    "test": hparams["test_annotation"],
}
hparams["dataloader_options"]["shuffle"] = False
for dataset in data_info:
    audio_files = glob.glob(os.path.join("file_path", dataset, "*.wav"))
    datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_csv(
        csv_path=data_info[dataset],
        replacements={"file_path": audio_files},
        dynamic_items=[audio_pipeline, label_pipeline],
        output_keys=["id", "sig", "spk_id_encoded"],
    )

Versions

No response

Relevant log output

No response

Additional context

No response

Start by comparing templates/speaker_id/train.py and train.yaml, especially the prepare_mini_librispeech call, dataio_prep, annotation paths, and save configuration. Run the reported training command with the supplied WAV files and JSON annotations, then verify that the configured save location contains the trained model artifacts.