nvidia/stt_da_fastconformer_rnnt_large

NVIDIA FastConformer-RNNT Large (Danish)

stt_da_fastconformer_rnnt_large

Description:

stt_da_fastconformer_rnnt_large is a 110-million-parameter automatic speech recognition (ASR) model trained to transcribe Danish speech.

This model uses the FastConformer architecture trained with RNN-T (RNN-Transducer) loss.

For more information, refer to the NeMo documentation.

This model is ready for commercial/non-commercial use.

License/Terms of Use:

GOVERNING TERMS: Use of this model is governed by the NVIDIA Open Model License Agreement (found at https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/).

Deployment Geography:

Global

Use Case:

Performing speech-to-text transcription in Danish language

Release Date:

Huggingface 10/15/25 via https://huggingface.co/nvidia/stt_da_fastconformer_rnnt_large/tree/main

References:

[1] Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition

[2] NVIDIA NeMo Toolkit

Model Architecture:

Architecture Type: Transducer model
Network Architecture: FastConformer encoder and LSTM prediction network

Input

Input Type(s): Audio
Input Format(s): 16000 Hz mono-channel
Input Parameters: 1D

Output

Output Type(s): Text Output Format: string Output Parameters: 1D Other Properties Related to Output: Not applicable

This model provides transcribed speech as a string for a given audio with Danish speech.

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration:

• NeMo-1.20
  

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Ampere
  
• NVIDIA Blackwell
  
• NVIDIA Hopper
  

Preferred Operating System(s):

Linux

Model Version:

1.0

Training and Evaluation Datasets:

The total size: ~10000 hours
Total number of datasets: 3

Training datasets

Link: CoRal (read aloud part)

  - Data Collection Method by Dataset: Human
  - Labeling Method by Dataset: Human
  - Properties: read aloud Danish speech

Link: Mozilla Common Voice v17

  - Data Collection Method by Dataset: Human
  - Labeling Method by Dataset: Automated
  - Properties: sentences recorded by different people

Link: Granary Danish a cleaned version of Danish part of MOSEL

  - Data Collection Method by Dataset: Human
  - Labeling Method by Dataset: Automated
  - Properties: EU Parlament sessions

Evaluation Datasets

Link: CoRal (read aloud part)

  - Data Collection Method by Dataset: Human
  - Labeling Method by Dataset: Human
  - Properties: read aloud Danish speech

Link: Mozilla Common Voice v17

  - Data Collection Method by Dataset: Human
  - Labeling Method by Dataset: Automated
  - Properties: sentences recorded by different people

Link: FLEURS

  - Data Collection Method by Dataset: Human
  - Labeling Method by Dataset: Human
  - Properties: sentences recorded by different people

Performance

The table below summarizes the WER (%) of this model using RNN-T decoder with greedy decoding:

Dataset	WER (%)
CoRal Test	10.7
MCV v17	8.8
Google FLEURS	10.6

These WER scores were obtained using greedy decoding without an external language model.

Inference

Acceleration Engine: Not Applicable Test Hardware: NVIDIA RTX 6000

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Please report security vulnerabilities or NVIDIA AI Concerns here.

How to Use this Model

To train, fine-tune or play with the model you will need to install NVIDIA NeMo. We recommend you install it after you've installed latest PyTorch version.

pip install nemo_toolkit['all']

Automatically instantiate the model

import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.ASRModel.from_pretrained(model_name="nvidia/stt_da_fastconformer_rnnt_large")

Transcribing using Python

Assuming you have an audio file with Danish speech simply do:

output = asr_model.transcribe(['danish_audio.wav'])
print(output[0])