🚀 Wav2Vec2-Large-100k-VoxPopuli-Català
This model is a fine - tuned version of facebook/wav2vec2-large-100k-voxpopuli on the Catalan language, aiming to provide high - quality automatic speech recognition for Catalan.
🚀 Quick Start
⚠️NOTICE⚠️: THIS MODEL HAS BEEN MOVED TO THE FOLLOWING URL:
https://huggingface.co/softcatala/wav2vec2-large-100k-voxpopuli-catala
This model is fine - tuned on the Common Voice and ParlamentParla datasets for the Catalan language.
⚠️ Important Note
The split train/dev/test used does not fully map with the CommonVoice 6.1 dataset. A custom split was used combining both the CommonVoice and ParlamentParla dataset and can be found here. Evaluating on the CV test dataset will produce a biased WER as 1144 audio files of that dataset were used in training/evaluation of this model. WER was calculated using this test.csv which was not seen by the model during training/evaluation.
💡 Usage Tip
When using this model, make sure that your speech input is sampled at 16kHz.
✨ Features
- Multidataset Training: Trained on multiple datasets including Common Voice and ParlamentParla to enhance the model's generalization ability.
- High - Quality Speech Recognition: Achieves relatively low word error rates on various Catalan speech datasets.
📦 Installation
No specific installation steps are provided in the original document.
💻 Usage Examples
Basic Usage
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "ca", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("ccoreilly/wav2vec2-large-100k-voxpopuli-catala")
model = Wav2Vec2ForCTC.from_pretrained("ccoreilly/wav2vec2-large-100k-voxpopuli-catala")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def speech_file_to_array_fn(batch):
speech_array, sampling_rate = torchaudio.load(batch["path"])
batch["speech"] = resampler(speech_array).squeeze().numpy()
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
predicted_ids = torch.argmax(logits, dim=-1)
print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset["sentence"][:2])
📚 Documentation
Model Information
Property |
Details |
Language |
Catalan |
Datasets |
Common Voice, ParlamentParla |
Metrics |
WER (Word Error Rate) |
Tags |
audio, automatic - speech - recognition, speech, speech - to - text |
License |
apache - 2.0 |
Model Index
- Name: Catalan VoxPopuli Wav2Vec2 Large
- Results:
- Task:
- Name: Speech Recognition
- Type: automatic - speech - recognition
- Datasets:
- Name: Common Voice ca
- Type: common_voice
- Args: ca
- Name: ParlamentParla
- URL: https://www.openslr.org/59/
- Metrics:
- Name: Test WER
- Type: wer
- Value: 5.98
- Name: Google Crowsourced Corpus WER
- Type: wer
- Value: 12.14
- Name: Audiobook “La llegenda de Sant Jordi” WER
- Type: wer
- Value: 12.02
Results
Word error rate was evaluated on the following datasets unseen by the model:
📄 License
This model is licensed under the apache - 2.0 license.