language: en
datasets:
- librispeech_asr
tags:
- speech
license: apache-2.0
Wav2Vec2-Large-960h-Lv60 + Self-Training
The large model pretrained and fine-tuned on 960 hours of Libri-Light and Librispeech. Model was trained with Self-Training objective
Authors: Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli
Abstract
We show for the first time that learning powerful representations from speech audio alone followed by fine-tuning on transcribed speech can outperform the best semi-supervised methods while being conceptually simpler. wav2vec 2.0 masks the speech input in the latent space and solves a contrastive task defined over a quantization of the latent representations which are jointly learned. Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/other test sets. When lowering the amount of labeled data to one hour, wav2vec 2.0 outperforms the previous state of the art on the 100 hour subset while using 100 times less labeled data. Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 4.8/8.2 WER. This demonstrates the feasibility of speech recognition with limited amounts of labeled data.
The original model can be found under https://github.com/pytorch/fairseq/tree/master/examples/wav2vec#wav2vec-20.
Usage
To transcribe audio files the model can be used as a standalone acoustic model as follows:
from transformers import Wav2Vec2Tokenizer, Wav2Vec2Model
from datasets import load_dataset
import soundfile as sf
import torch
# load model and tokenizer
tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-large-960h-lv60-self")
model = Wav2Vec2ForMaskedLM.from_pretrained("facebook/wav2vec2-large-960h-lv60-self")
# define function to read in sound file
def map_to_array(batch):
speech, _ = sf.read(batch["file"])
batch["speech"] = speech
return batch
# load dummy dataset and read soundfiles
ds = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation")
ds = ds.map(map_to_array)
# tokenize
input_values = tokenizer(ds["speech"][:2], return_tensors="pt", padding="longest").input_values # Batch size 1
# retrieve logits
logits = model(input_values).logits
# take argmax and decode
predicted_ids = torch.argmax(logits, dim=-1)
transcription = tokenizer.batch_decode(predicted_ids)