File size: 2,987 Bytes
e370a4d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 |
---
language: en
datasets:
- superb
tags:
- speech
- audio
- hubert
- s3prl
license: apache-2.0
---
# Hubert-Base for Intent Classification
## Model description
This is a ported version of [S3PRL's Hubert for the SUPERB Intent Classification task](https://github.com/s3prl/s3prl/tree/master/s3prl/downstream/fluent_commands).
The base model is [hubert-base-ls960](https://huggingface.co/facebook/hubert-base-ls960).
It 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](https://arxiv.org/abs/2105.01051)
## Task description
Intent Classification (IC) classifies utterances into predefined classes to determine the intent of
speakers. SUPERB uses the
[Fluent Speech Commands](https://fluent.ai/fluent-speech-commands-a-dataset-for-spoken-language-understanding-research/)
dataset, where each utterance is tagged with three intent labels: **action**, **object**, and **location**.
For the original model's training and evaluation instructions refer to the
[S3PRL downstream task README](https://github.com/s3prl/s3prl/tree/master/s3prl/downstream#ic-intent-classification---fluent-speech-commands).
## Usage examples
You can use the model directly like so:
```python
import torch
import numpy as np
from datasets import load_dataset
from transformers import HubertForSequenceClassification, Wav2Vec2FeatureExtractor
# TODO: replace with the official superb dataset
superb_ks = load_dataset("anton-l/superb_dummy", "ic", split="test")
model = HubertForSequenceClassification.from_pretrained("superb/hubert-base-superb-ic")
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("superb/hubert-base-superb-ic")
audio = np.array(superb_ks[0]["speech"])
# compute attention masks and normalize the waveform if needed
inputs = feature_extractor(audio, sampling_rate=16_000, return_tensors="pt")
logits = model(**inputs).logits[0]
action_ids = torch.argmax(logits[:, :6], dim=-1).tolist()
action_labels = [model.config.id2label[_id] for _id in action_ids]
object_ids = torch.argmax(logits[:, 6:20], dim=-1).tolist()
object_labels = [model.config.id2label[_id + 6] for _id in object_ids]
location_ids = torch.argmax(logits[:, 20:24], dim=-1).tolist()
location_labels = [model.config.id2label[_id + 20] for _id in location_ids]
```
## Eval results
The evaluation metric is accuracy.
| | `s3prl` | `transformers` |
|------|---------|----------------|
|`test`| TBA | TBA |
### BibTeX entry and citation info
```bibtex
@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}
}
``` |