metadata

language: en
datasets:
  - superb
tags:
  - speech
  - audio
  - wav2vec2
  - audio-classification
license: apache-2.0
widget:
  - example_title: Speech Commands "down"
    src: https://cdn-media.huggingface.co/speech_samples/keyword_spotting_down.wav
  - example_title: Speech Commands "go"
    src: https://cdn-media.huggingface.co/speech_samples/keyword_spotting_go.wav

Wav2Vec2-Large for Keyword Spotting

Model description

This is a ported version of S3PRL's Wav2Vec2 for the SUPERB Keyword Spotting task.

The base model is wav2vec2-large-lv60, which is pretrained on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16Khz.

For more information refer to SUPERB: Speech processing Universal PERformance Benchmark

Task and dataset description

Keyword Spotting (KS) detects preregistered keywords by classifying utterances into a predefined set of words. The task is usually performed on-device for the fast response time. Thus, accuracy, model size, and inference time are all crucial. SUPERB uses the widely used Speech Commands dataset v1.0 for the task. The dataset consists of ten classes of keywords, a class for silence, and an unknown class to include the false positive.

For the original model's training and evaluation instructions refer to the S3PRL downstream task README.

Usage examples

You can use the model via the Audio Classification pipeline:

from datasets import load_dataset
from transformers import pipeline

dataset = load_dataset("anton-l/superb_demo", "ks", split="test")

classifier = pipeline("audio-classification", model="superb/wav2vec2-large-superb-ks")
labels = classifier(dataset[0]["file"], top_k=5)

Or use the model directly:

import torch
from datasets import load_dataset
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor
from torchaudio.sox_effects import apply_effects_file

effects = [["channels", "1"], ["rate", "16000"], ["gain", "-3.0"]]
def map_to_array(example):
    speech, _ = apply_effects_file(example["file"], effects)
    example["speech"] = speech.squeeze(0).numpy()
    return example

# load a demo dataset and read audio files
dataset = load_dataset("anton-l/superb_demo", "ks", split="test")
dataset = dataset.map(map_to_array)

model = Wav2Vec2ForSequenceClassification.from_pretrained("superb/wav2vec2-large-superb-ks")
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("superb/wav2vec2-large-superb-ks")

# compute attention masks and normalize the waveform if needed
inputs = feature_extractor(dataset[:4]["speech"], sampling_rate=16000, padding=True, return_tensors="pt")

logits = model(**inputs).logits
predicted_ids = torch.argmax(logits, dim=-1)
labels = [model.config.id2label[_id] for _id in predicted_ids.tolist()]

Eval results

The evaluation metric is accuracy.

	s3prl	transformers
test	`0.9666`	`N/A`

BibTeX entry and citation info

@article{yang2021superb,
  title={SUPERB: Speech processing Universal PERformance Benchmark},
  author={Yang, Shu-wen and Chi, Po-Han and Chuang, Yung-Sung and Lai, Cheng-I Jeff and Lakhotia, Kushal and Lin, Yist Y and Liu, Andy T and Shi, Jiatong and Chang, Xuankai and Lin, Guan-Ting and others},
  journal={arXiv preprint arXiv:2105.01051},
  year={2021}
}