Model creates duplicate transcriptions

[ceyxasm]

import nemo.collections.asr as nemo_asr
import sys
MODEL_PATH = '/home/bubu/attention-tag/customs/asr-onprem/parakeet-tdt_ctc-110m.nemo'

asr_model = nemo_asr.models.ASRModel.restore_from(MODEL_PATH)
transcriptions = asr_model.transcribe([sys.argv[1]])

With a single audio wav file, transcriptions consist of a tuple consisting of two transcriptions which are duplicate of each other.
PS: this issue is a copy of hugging face discussion: https://huggingface.co/nvidia/parakeet-tdt_ctc-110m/discussions/2

[nithinraok]

Hi, thanks for the issue,
We updated our signature with latest release (2.2)
and we updated the card accordingly: https://huggingface.co/nvidia/parakeet-tdt_ctc-110m#how-to-use-this-model

please check and let us know.

import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.ASRModel.from_pretrained("nvidia/parakeet-tdt_ctc-110m")
transcriptions = asr_model.transcribe(['<file_path>'])
print(transcriptions[0].text)

[ceyxasm]

Hey but still the need to do print(transcriptions[0].text) is fishy right?
And if you were to compare transcriptions[0].text==transcriptions[1].text, you would get true.
My concern was this redundancy. Not a breaking bug but a bug nonetheless

[nithinraok]

why do you get ranscriptions[1].text when you pass only one audio file? pls provide code to replicate your issue

[ceyxasm]

import nemo.collections.asr as nemo_asr
import sys
import time
MODEL_PATH = '/home/bubu/attention-tag/customs/asr-onprem/parakeet-tdt_ctc-110m.nemo'

asr_model = nemo_asr.models.ASRModel.restore_from(MODEL_PATH)
wav_file_path = sys.argv[1]
st = time.time()
transcriptions = asr_model.transcribe(wav_file_path)
et = time.time() - st
flat_transcriptions = [item for sublist in transcriptions for item in sublist]

print(et, len(transcriptions))
print(transcriptions[1])
with open('trans.txt', 'w') as f:
    f.write("\n".join(flat_transcriptions))

this transcription[1] is same as transcription[0]