Sortformer does not pick up audio duration from the manifest

[orech]

Describe the bug

Hi team,

I was experimenting with streaming sortformer and discovered that it was not picking up duration parameter from the input manifest. Based on HF instructions the model expects a manifest file, where offset and duration of the slice that needs to be diarized are set as:

# Example of a line in `multispeaker_manifest.json`
{
    "audio_filepath": "/path/to/multispeaker_audio1.wav",  # path to the input audio file 
    "offset": 0, # offset (start) time of the input audio
    "duration": 600,  # duration of the audio, can be set to `null` if using NeMo main branch
}
{
    "audio_filepath": "/path/to/multispeaker_audio2.wav",  
    "offset": 900,
    "duration": 580,  
}

https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2

I suspect that could be because at this line the code ignores all fields in the manifest except for audio_filepath:

https://github.com/NVIDIA-NeMo/NeMo/blob/main/nemo/collections/asr/parts/mixins/diarization.py#L373

Later _diarize_input_manifest_processing iterates over audio files, where each entry is a string (as this is the only field preserved at the previous step) and populates other entries with:

entry = {'audio_filepath': audio_file, 'duration': 100000, 'text': ''}

https://github.com/NVIDIA-NeMo/NeMo/blob/main/nemo/collections/asr/parts/mixins/diarization.py#L405

I might be reading the code wrong, please help me figure out if my inputs to the model are invalid

[tango4j]

Hi. This seems like a valid issue.
We will check and queue this issue to fix it so that it reflects it and only work on [offset, offset+duration].
Thanks for raising this issue.