wav2vec2-xls-r-parlaspeech-hr-lm

This model for Croatian ASR is based on the facebook/wav2vec2-xls-r-300m model and was fine-tuned with 300 hours of recordings and transcripts from the ASR Croatian parliament dataset ParlaSpeech-HR v1.0.

If you use this model, please cite the following paper:

@inproceedings{ljubevsic2022parlaspeech,
  title={ParlaSpeech-HR-a Freely Available ASR Dataset for Croatian Bootstrapped from the ParlaMint Corpus},
  author={Ljube{\v{s}}i{\'c}, Nikola and Kor{\v{z}}inek, Danijel and Rupnik, Peter and Jazbec, Ivo-Pavao},
  booktitle={Proceedings of the Workshop ParlaCLARIN III within the 13th Language Resources and Evaluation Conference},
  pages={111--116},
  year={2022},
  url={http://www.lrec-conf.org/proceedings/lrec2022/workshops/ParlaCLARINIII/pdf/2022.parlaclariniii-1.16.pdf}
}

Metrics

Evaluation is performed on the dev and test portions of the ParlaSpeech-HR v1.0 dataset.

split	CER	WER
dev	0.0448	0.1129
test	0.0363	0.0985

There are multiple models available, and in terms of CER and WER, the best-performing model is wav2vec2-large-slavic-parlaspeech-hr-lm.

Usage in transformers

Tested with transformers==4.18.0, torch==1.11.0, and SoundFile==0.10.3.post1.

from transformers import Wav2Vec2ProcessorWithLM, Wav2Vec2ForCTC
import soundfile as sf
import torch
import os
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# load model and tokenizer
processor = Wav2Vec2ProcessorWithLM.from_pretrained(
    "classla/wav2vec2-xls-r-parlaspeech-hr-lm")
model = Wav2Vec2ForCTC.from_pretrained("classla/wav2vec2-xls-r-parlaspeech-hr-lm")
# download the example wav files:
os.system("wget https://huggingface.co/classla/wav2vec2-large-slavic-parlaspeech-hr/raw/main/00020570a.flac.wav")
# read the wav file 
speech, sample_rate = sf.read("00020570a.flac.wav")
input_values = processor(speech, sampling_rate=sample_rate, return_tensors="pt").input_values.cuda()
inputs = processor(speech, sampling_rate=sample_rate, return_tensors="pt")
with torch.no_grad():
    logits = model(**inputs).logits
transcription = processor.batch_decode(logits.numpy()).text[0]

# remove the raw wav file
os.system("rm 00020570a.flac.wav")
transcription

# transcription: 'velik broj poslovnih subjekata posluje sa minusom velik dio'

Training hyperparameters

In fine-tuning, the following arguments were used:

arg	value
per_device_train_batch_size	16
gradient_accumulation_steps	4
num_train_epochs	8
learning_rate	3e-4
warmup_steps	500