Fine-tuned XLS-R 1B model for speech recognition in Portuguese
Fine-tuned facebook/wav2vec2-xls-r-1b on Portuguese using the train and validation splits of Common Voice 8.0, CORAA, Multilingual TEDx, and Multilingual LibriSpeech. 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 :)
Usage
Using the HuggingSound library:
from huggingsound import SpeechRecognitionModel
model = SpeechRecognitionModel("jonatasgrosman/wav2vec2-xls-r-1b-portuguese")
audio_paths = ["/path/to/file.mp3", "/path/to/another_file.wav"]
transcriptions = model.transcribe(audio_paths)
Writing your own inference script:
import torch
import librosa
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
LANG_ID = "pt"
MODEL_ID = "jonatasgrosman/wav2vec2-xls-r-1b-portuguese"
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)
Evaluation Commands
- To evaluate on
mozilla-foundation/common_voice_8_0
with splittest
python eval.py --model_id jonatasgrosman/wav2vec2-xls-r-1b-portuguese --dataset mozilla-foundation/common_voice_8_0 --config pt --split test
- To evaluate on
speech-recognition-community-v2/dev_data
python eval.py --model_id jonatasgrosman/wav2vec2-xls-r-1b-portuguese --dataset speech-recognition-community-v2/dev_data --config pt --split validation --chunk_length_s 5.0 --stride_length_s 1.0
Citation
If you want to cite this model you can use this:
@misc{grosman2021xlsr-1b-portuguese,
title={Fine-tuned {XLS-R} 1{B} model for speech recognition in {P}ortuguese},
author={Grosman, Jonatas},
howpublished={\url{https://huggingface.co/jonatasgrosman/wav2vec2-xls-r-1b-portuguese}},
year={2022}
}
- Downloads last month
- 635
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 medtalkai/wav2vec_kenlm5
Evaluation results
- Test WER on Common Voice 8self-reported8.700
- Test CER on Common Voice 8self-reported2.550
- Test WER (+LM) on Common Voice 8self-reported6.040
- Test CER (+LM) on Common Voice 8self-reported1.980
- Dev WER on Robust Speech Event - Dev Dataself-reported24.230
- Dev CER on Robust Speech Event - Dev Dataself-reported11.300
- Dev WER (+LM) on Robust Speech Event - Dev Dataself-reported19.410
- Dev CER (+LM) on Robust Speech Event - Dev Dataself-reported10.190
- Test WER on Robust Speech Event - Test Dataself-reported18.800