YAML Metadata Error: "language[0]" must only contain lowercase characters
YAML Metadata Error: "language[0]" with value "nb-NO" is not valid. It must be an ISO 639-1, 639-2 or 639-3 code (two/three letters), or a special value like "code", "multilingual". If you want to use BCP-47 identifiers, you can specify them in language_bcp47.
YAML Metadata Error: "tags[2]" must be a string

XLS-R-300M-LM - Norwegian

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the Norwegian NPSC dataset.

Scores without Language Model

Without using a language model, it achieves the following scores on the NPSC Eval set It achieves the following results on the evaluation set without a language model:

  • WER: 0.2110
  • CER: 0.0622

Scores with Language Model

A 5-gram KenLM was added to boost the models performance. The language model was created on a corpus mainly consisting of online newspapers, public reports and Wikipedia data. After this we are getting these values.

  • WER: 0.1540
  • CER: 0.0548

Team

The model is developed by Rolv-Arild Braaten, Per Egil Kummervold, Andre Kåsen, Javier de la Rosa, Per Erik Solberg, and Freddy Wetjen. Name in alphabetic order.

Model description

This current version is based on checkpoint 8500 of NbAiLab/wav2vec2-xlsr-300M-NPSC-OH.

Intended uses & limitations

Demo version only. The model will be updated later this week.

Training and evaluation data

The model is trained and evaluated on NPSC. Unfortunately there is no Norwegian test data in Common Voice, and currently the model is only evaluated on the validation set of NPSC..

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 7.5e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 2000
  • num_epochs: 30.0 (But interrupted after 8500 steps, approx 6 epochs)
  • mixed_precision_training: Native AMP
Downloads last month
16
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train NbAiLab/wav2vec2-xlsr-300M-NPSC-LM

Evaluation results