Wav2Vec2-Large-100h-Lv60 + Self-Training

This is a direct state_dict transfer from fairseq to huggingface, the weights are identical

Facebook's Wav2Vec2

The large model pretrained and fine-tuned on 100 hours of Libri-Light and Librispeech on 16kHz sampled speech audio. Model was trained with Self-Training objective. When using the model make sure that your speech input is also sampled at 16Khz.

Paper

Authors: Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli

Abstract

They 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 Wav2Vec2Processor, Wav2Vec2ForCTC
 from datasets import load_dataset
 import torch
 
 # load model and processor
 processor = Wav2Vec2Processor.from_pretrained("Splend1dchan/wav2vec2-large-100h-lv60-self")
 model = Wav2Vec2ForCTC.from_pretrained("Splend1dchan/wav2vec2-large-100h-lv60-self")
     
 # load dummy dataset and read soundfiles
 ds = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation")
 
 # tokenize
 input_values = processor(ds[0]["audio"]["array"], return_tensors="pt", padding="longest").input_values
 
 # retrieve logits
 logits = model(input_values).logits
 
 # take argmax and decode
 predicted_ids = torch.argmax(logits, dim=-1)
 transcription = processor.batch_decode(predicted_ids)

Evaluation

This code snippet shows how to evaluate facebook's Splend1dchan/wav2vec2-large-100h-lv60-self on LibriSpeech's "clean" and "other" test data.

from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import torch
from jiwer import wer
librispeech_eval = load_dataset("librispeech_asr", "clean", split="test")
model = Wav2Vec2ForCTC.from_pretrained("Splend1dchan/wav2vec2-large-100h-lv60-self").to("cuda")
processor = Wav2Vec2Processor.from_pretrained("Splend1dchan/wav2vec2-large-100h-lv60-self")
def map_to_pred(batch):
    inputs = processor(batch["audio"]["array"], return_tensors="pt", padding="longest")
    input_values = inputs.input_values.to("cuda")
    attention_mask = inputs.attention_mask.to("cuda")
    
    with torch.no_grad():
        logits = model(input_values, attention_mask=attention_mask).logits
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.batch_decode(predicted_ids)
    batch["transcription"] = transcription
    return batch
result = librispeech_eval.map(map_to_pred, remove_columns=["speech"])
print("WER:", wer(result["text"], result["transcription"]))