wav2vec2-100h

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 Wav2Vec2Processor, Wav2Vec2ForCTC
 from datasets import load_dataset
 import soundfile as sf
 import torch

 # load model and processor
 processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-100h")
 model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-100h")

 # 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 = processor(ds[0]["audio"]["array"], 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 = processor.batch_decode(predicted_ids)

Evaluation

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

from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import soundfile as sf
import torch
from jiwer import wer

librispeech_eval = load_dataset("librispeech_asr", "clean", split="test")

model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-100h").to("cuda")
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-100h")

def map_to_pred(batch):
    input_values = processor(batch["audio"]["array"], return_tensors="pt", padding="longest").input_values
    with torch.no_grad():
        logits = model(input_values.to("cuda")).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, batched=True, batch_size=1, remove_columns=["speech"])

print("WER:", wer(result["text"], result["transcription"]))

Result (WER):