Wav2vec2 Xls R 1b Italian

🚀 Fine-tuned XLS-R 1B model for speech recognition in Italian

This is a fine-tuned model based on facebook/wav2vec2-xls-r-1b for Italian speech recognition, which can effectively convert Italian speech into text.

🚀 Quick Start

This model is fine-tuned on the facebook/wav2vec2-xls-r-1b using the train and validation splits of Common Voice 8.0, Multilingual TEDx, Multilingual LibriSpeech, and Voxpopuli. When using this model, make sure that your speech input is sampled at 16kHz.

This model has been fine-tuned by the HuggingSound tool, and thanks to the GPU credits generously given by the OVHcloud :)

✨ Features

• Multilingual Adaptability: Based on the XLS - R Wav2Vec2 architecture, it shows good adaptability in Italian speech recognition.
• High - Quality Training: Trained on multiple high - quality datasets, including Common Voice 8.0, Multilingual TEDx, Multilingual LibriSpeech, and Voxpopuli.
• Performance Metrics: Achieved good results in WER (Word Error Rate) and CER (Character Error Rate) on different datasets.

📦 Installation

There is no specific installation content provided in the original document. If you want to use this model, you can install the necessary libraries according to the usage examples, such as transformers, torch, librosa, datasets etc.

💻 Usage Examples

Basic Usage

Using the HuggingSound library:

from huggingsound import SpeechRecognitionModel

model = SpeechRecognitionModel("jonatasgrosman/wav2vec2-xls-r-1b-italian")
audio_paths = ["/path/to/file.mp3", "/path/to/another_file.wav"]

transcriptions = model.transcribe(audio_paths)

Advanced Usage

Writing your own inference script:

import torch
import librosa
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor

LANG_ID = "it"
MODEL_ID = "jonatasgrosman/wav2vec2-xls-r-1b-italian"
SAMPLES = 10

test_dataset = load_dataset("common_voice", LANG_ID, split=f"test[:{SAMPLES}]")

processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)
model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)

# Preprocessing the datasets.
# We need to read the audio files as arrays
def speech_file_to_array_fn(batch):
    speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000)
    batch["speech"] = speech_array
    batch["sentence"] = batch["sentence"].upper()
    return batch

test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"], 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)
predicted_sentences = processor.batch_decode(predicted_ids)

📚 Documentation

Evaluation Commands

1. To evaluate on mozilla-foundation/common_voice_8_0 with split test

python eval.py --model_id jonatasgrosman/wav2vec2-xls-r-1b-italian --dataset mozilla-foundation/common_voice_8_0 --config it --split test

2. To evaluate on speech-recognition-community-v2/dev_data

python eval.py --model_id jonatasgrosman/wav2vec2-xls-r-1b-italian --dataset speech-recognition-community-v2/dev_data --config it --split validation --chunk_length_s 5.0 --stride_length_s 1.0

Model Index

📄 License

This model is licensed under the Apache - 2.0 license.

📚 Citation

If you want to cite this model you can use this:

@misc{grosman2021xlsr-1b-italian,
  title={Fine-tuned {XLS-R} 1{B} model for speech recognition in {I}talian},
  author={Grosman, Jonatas},
  howpublished={\url{https://huggingface.co/jonatasgrosman/wav2vec2-xls-r-1b-italian}},
  year={2022}
}