nvidia

/

parakeet-ctc-0.6b-Vietnamese

like 78

Follow

NVIDIA 51.3k

Hi @tridm , can you try download model from huggingface and restore checkpoint from local?

Hi @tridm , has your problem been solved? If it solved, could you close this discussion?

Hi @hoanguyen0705
I'm running the script on macOS and encounter a FileNotFoundError for model_config.yaml.

Environment:

• OS: macOS
• NeMo: 2.6.1

Script:

import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.ASRModel.from_pretrained("nvidia/parakeet-ctc-0.6b-Vietnamese")

The model nvidia/parakeet-ctc-0.6b-Vietnamese is successfully downloaded from HuggingFace,
but during restoration NeMo raises:

FileNotFoundError: No such file or directory: model_config.yaml

Is this a macOS-related issue, a HuggingFace loading issue, or is the model missing the config file?
Any guidance would be appreciated. Thanks!

I can't run this model. After I install NeMo using the command (!pip install -U nemo_toolkit['asr']), Colab automatically restarts. Please help me. Thank you.

Hi @PhucTinh @ltlonggg , you can try asr_model = nemo_asr.models.ASRModel.from_pretrained("nvidia/parakeet-ctc-0.6b-vi") as the acoustic model.
For the best results, I suggest downloading the model from Hugging Face to local storage so you can get the language model and follow the tutorial for inference with an n-gram and a lexicon.
You can follow this tutorial to download: https://huggingface.co/docs/huggingface_hub/en/guides/download

Thank, I see 'parakeet-ctc-0.6b-vi.nemo' file and many other file in Files but i don't know how to run those file. Can you guide me ?
Thank you so much.
(I've found the way to do it)

And currently it's experiencing library conflicts. Do you have any way to fix this problem?

Thank you

Hi @hoanguyen0705
Thanks! I followed your suggestion and was able to run inference successfully.

Hi @ltlonggg
I used the following script and was able to run inference successfully:

import librosa
import soundfile as sf
import os

def preprocess_audio(audio_path):
    target_sr = 16000
    y, sr = librosa.load(audio_path, sr=target_sr, mono=True)
    
    processed_path = "temp_mono_16k.wav"
    sf.write(processed_path, y, target_sr)
    return processed_path

import nemo.collections.asr as nemo_asr

model_path = "nvidia_parakeet_ctc_0.6b_vietnamese/parakeet-ctc-0.6b-vi.nemo"

asr_model = nemo_asr.models.ASRModel.restore_from(restore_path=model_path)

raw_audio = "audio.wav"
ready_audio = preprocess_audio(raw_audio)

transcriptions = asr_model.transcribe([ready_audio])
print(f"Kết quả: {transcriptions[0].text}")

if os.path.exists(ready_audio):
    os.remove(ready_audio)

hi @PhucTinh , thank you for this code snippet. But while install 'nemo_asr' library, I see that the libraries are in conflict with each other beacause I am running in Colab, therefore, I would like to ask how to fix that conflict error, and by the way, what platform are you running your code on ?
Thank you so much

@ltlonggg
Hi, I’m running this code locally in a Jupyter Notebook inside Cursor.

My tries :
Ran successfully within nemo_toolkit === 2.5.0
Ran failed within latest nemo_toolkit 2.6.0+

Error log of v2.6.0

[NeMo I 2026-02-02 17:01:14 nemo_logging:393] PADDING: 0
[NeMo I 2026-02-02 17:01:15 nemo_logging:393] Model EncDecCTCModelBPE was successfully restored from /Users/khanhicetea/.cache/torch/NeMo/NeMo_2.6.1/hf_hub_cache/nvidia/parakeet-ctc-0.6b-Vietnamese/ad1356a8c624a67221ed80e50d009746/parakeet-ctc-0.6b-vi.nemo.
[NeMo W 2026-02-02 17:01:16 nemo_logging:405] The following configuration keys are ignored by Lhotse dataloader: use_start_end_token
[NeMo W 2026-02-02 17:01:16 nemo_logging:405] You are using a non-tarred dataset and requested tokenization during data sampling (pretokenize=True). This will cause the tokenization to happen in the main (GPU) process,possibly impacting the training speed if your tokenizer is very large.If the impact is noticable, set pretokenize=False in dataloader config.(note: that will disable token-per-second filtering and 2D bucketing features)
Traceback (most recent call last):
  File "/Users/khanhicetea/Code/labs/nemo_paraket/abc.py", line 25, in <module>
    transcriptions = asr_model.transcribe([ready_audio])
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/asr/models/ctc_models.py", line 181, in transcribe
    return super().transcribe(
           ^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/asr/parts/mixins/transcription.py", line 270, in transcribe
    for processed_outputs in generator:
                             ^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/asr/parts/mixins/transcription.py", line 356, in transcribe_generator
    dataloader = self._transcribe_input_processing(audio, transcribe_cfg)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/asr/parts/mixins/transcription.py", line 473, in _transcribe_input_processing
    temp_dataloader = self._setup_transcribe_dataloader(ds_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/asr/models/ctc_bpe_models.py", line 207, in _setup_transcribe_dataloader
    temporary_datalayer = self._setup_dataloader_from_config(config=DictConfig(dl_config))
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/asr/models/ctc_bpe_models.py", line 98, in _setup_dataloader_from_config
    return get_lhotse_dataloader_from_config(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/common/data/lhotse/dataloader.py", line 266, in get_lhotse_dataloader_from_config
    return get_lhotse_dataloader_from_single_config(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/common/data/lhotse/dataloader.py", line 306, in get_lhotse_dataloader_from_single_config
    sampler, use_iterable_dataset = get_lhotse_sampler_from_config(
                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/nemo/collections/common/data/lhotse/dataloader.py", line 595, in get_lhotse_sampler_from_config
    sampler = DynamicCutSampler(
              ^^^^^^^^^^^^^^^^^^
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/lhotse/dataset/sampling/dynamic.py", line 120, in __init__
    super().__init__(
  File "/Users/khanhicetea/Code/labs/nemo_paraket/.venv/lib/python3.12/site-packages/lhotse/dataset/sampling/base.py", line 76, in __init__
    super().__init__(
TypeError: object.__init__() takes exactly one argument (the instance to initialize)