Wav2vec2 Large Voxpopuli Sv Swedish

🚀 Wav2vec 2.0 large-voxpopuli-sv-swedish

An additionally pretrained and finetuned model for Swedish automatic speech recognition, leveraging Swedish radio broadcasts, NST, and Common Voice data.

⚠️ Important Note

this model performs better and has a less restrictive license.

This is an additionally pretrained and finetuned version of Facebook's VoxPopuli-sv large model. It uses Swedish radio broadcasts, NST, and Common Voice data. Evaluation without a language model shows the following results: The WER for the NST + Common Voice test set (2% of total sentences) is 3.95%. The WER for the Common Voice test set is 10.99% directly and 7.82% with a 4-gram language model.

💡 Usage Tip

When using this model, make sure that your speech input is sampled at 16kHz.

🚀 Quick Start

✨ Features

• Trained on multiple Swedish datasets including Common Voice and NST Swedish ASR Database.
• Evaluated using metrics like WER and CER.
• Can be used for automatic speech recognition tasks.

📦 Installation

No specific installation steps 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", "sv-SE", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("KBLab/wav2vec2-large-voxpopuli-sv-swedish")
model = Wav2Vec2ForCTC.from_pretrained("KBLab/wav2vec2-large-voxpopuli-sv-swedish")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
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])

🔧 Technical Details

This model has additionally pretrained on 1000h of Swedish local radio broadcasts, fine-tuned for 120000 updates on NST + CommonVoice and then for an additional 20000 updates on CommonVoice only. The additional fine-tuning on CommonVoice hurts performance on the NST+CommonVoice test set somewhat and, unsurprisingly, improves it on the CommonVoice test set. It seems to perform generally better though [citation needed].

📄 License

This model is licensed under cc-by-nc-4.0.

📚 Documentation