canary-1b not proccessing json files correctly about nemo

[titu1994]

Hmm can you try to see if you remove trailing new line in your manifest file, and also try to see if removing the json indent helps ?

[krishnacpuvvada]

return func(*args, **kwargs)
TypeError: EncDecMultiTaskModel.transcribe() got an unexpected keyword argument 'audio

can you verify that you are working with main branch of NeMo? the argument was updated from paths2audio_files to audio from r1.23 to main branch.

Also as a sanity check, can you try the following on main branch

trascript = canary_model.trascribe(audio=['path_to_file.wav']) # assuming path_to_file.wav is english audio.

[P15V]

@titu1994 I am experiencing much the same issue on my end. I tried as such and had no luck. I get the same JSON error as the one posted above.
when doing as suggested by removing the JSON indent and trailing lines. I get the following error:
"json.decoder.JSONDecodeError: Expecting property name enclosed in double quotes: line 1 column 171 (char 170)"

When directly fixing the JSON errors, I end up with this error that suggests it expects the JSON to have an 'answer' key somewhere?!? :

"-packages/nemo/collections/common/data/lhotse/nemo_adapters.py", line 84, in iter
text=data[self.text_field],
KeyError: 'answer'
Transcribing: 0it [00:00, ?it/s]"

I want to run ASR, so expecting an 'answer' key is wonky.

[krishnacpuvvada]

text=data[self.text_field],
KeyError: 'answer'

to fix this, please modify the input lines file to add 'answer' field.

{
  "audio_filepath": "PathRemovedDueToPersonalName",  
  "duration": 30.0,  
  "taskname": "asr",  
  "source_lang": "en", 
  "target_lang": "en", 
  "pnc": 'yes', 
  "answer": 'na',
}

also, we recently updated .transcribe signature, so if you are using main branch

transcript = canary_model.transcribe(paths2audio_files="/home/pjstimac/NvidiaCanaryTest/transcribe_manifest.json", batch_size=16)

should be updated to
transcript = canary_model.transcribe(audio="/home/pjstimac/NvidiaCanaryTest/transcribe_manifest.json", batch_size=16)

[P15V]

@krishnacpuvvada No luck, unfortunately. Tried that exact format(with JSON & JSONL) with the updated transcript variable. Yields this error
"
return func(*args, **kwargs)
TypeError: EncDecMultiTaskModel.transcribe() got an unexpected keyword argument 'audio
'"

[P15V]

@krishnacpuvvada I'll double-check and update again when I'm home. But I was following the Nvidia tutorial and git-cloning the repo just last night, so I'm assuming it's the latest as of yesterday evening.

The good news is I finally got basic inference working with three lines of code, and forgoing the JSON/JSONL manifest.

"
import nemo.collections.asr as nemo_asr
nemoasr_model = nemo_asr.models.EncDecCTCModelBPE.from_pretrained("nvidia/canary-1b")
nemoasr_model.transcribe(['AudioClipDirectly.wav'])

"

To accomplish the transcription of a directory of audio clips, I wrote some Python code that loops through the directory and outputs the transcription results as JSON for my local model comparison app.

[titu1994]

You can use this high level script to do inference with Canary as well - https://github.com/NVIDIA/NeMo/blob/main/examples/asr/transcribe_speech.py

[P15V]

@titu1994 I did try that as well per the Nvidia NeMo tutorial. Leading to the same exact issues I was running into elsewhere.