Canary examples not working

[icsy7867]

I am trying to get the Canary examples working.

I have a single .mp3 file in the folder. And trying to follow:
https://huggingface.co/nvidia/canary-1b

# load model
canary_model = EncDecMultiTaskModel.from_pretrained('nvidia/canary-1b')

# update dcode params
decode_cfg = canary_model.cfg.decoding
decode_cfg.beam.beam_size = 1
canary_model.change_decoding_strategy(decode_cfg)

predicted_text = canary_model.transcribe(
    audio_dir = "/tmp/gradio/a917afa6fde808c0cc5abf55cf0b14fd20ac189e/",
    batch_size=16,  # batch size to run the inference with
)

print(predicted_text)

Looks like this does not like "audio_dir" as the articles on huggingface suggest.

I also tried the example speech to text...

python ./NeMo/examples/asr/transcribe_speech.py pretrained_name="nvidia/canary-1b" dataset_manifest=/tmp/gradio/manifest.json

The manifest.json:

{"audio_filepath": "/tmp/gradio/a917afa6fde808c0cc5abf55cf0b14fd20ac189e/2006_0327_LB_CA_RedevBoard_JohnEastman.ogg.mp3","duration": "None","taskname": "asr","source_lang": "en","target_lang": "en","pnc": "yes","answer": "na"}

and trying again:

python ./NeMo/examples/asr/transcribe_speech.py pretrained_name="nvidia/canary-1b" dataset_manifest="/tmp/gradio/manifest.json"

Different error:

ImportError: cannot import name 'read_and_maybe_sort_manifest' from 'nemo.collections.asr.parts.utils.transcribe_utils' (/usr/local/lib/python3.10/site-packages/nemo/collections/asr/parts/utils/transcribe_utils.py)

I have also tried to use the directory method with the example script. However nothing seems to be working from the documentation.

[P15V]

Same issue here :(