The base Facebook's Wav2Vec2 model pretrained on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16Khz.

Task and dataset description

Intent Classification (IC) classifies utterances into predefined classes to determine the intent of speakers. The dataset used here is SLURP, where each utterance is tagged with two intent labels: action and scenario.

Usage examples

You can use the model directly in the following manner:

python

import torch
import librosa
from transformers import AutoModelForAudioClassification, AutoFeatureExtractor

## Load an audio file
audio_array, sr = librosa.load("path_to_audio.wav", sr=16000)

## Load model and feature extractor
model = AutoModelForAudioClassification.from_pretrained("alkiskoudounas/wav2vec2-base-slurp")
feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-base")

## Extract features
inputs = feature_extractor(audio_array.squeeze(), sampling_rate=feature_extractor.sampling_rate, padding=True, return_tensors="pt")

## Compute logits
logits = model(**inputs).logits

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-04
train_batch_size: 32
eval_batch_size: 32
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 128
optimizer: AdamW with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
warmup_steps: 3000
num_steps: 30000

Framework versions

Datasets 3.2.0
Pytorch 2.1.2
Tokenizers 0.20.3
Transformers 4.45.2

BibTeX entry and citation info

bibtex

@ARTICLE{koudounas2024taslp,
  author={Koudounas, Alkis and Pastor, Eliana and Attanasio, Giuseppe and Mazzia, Vittorio and Giollo, Manuel and Gueudre, Thomas and Reale, Elisa and Cagliero, Luca and Cumani, Sandro and de Alfaro, Luca and Baralis, Elena and Amberti, Daniele},
  journal={IEEE/ACM Transactions on Audio, Speech, and Language Processing}, 
  title={Towards Comprehensive Subgroup Performance Analysis in Speech Models}, 
  year={2024},
  volume={32},
  number={},
  pages={1468-1480},
  keywords={Analytical models;Task analysis;Metadata;Speech processing;Behavioral sciences;Itemsets;Speech;Speech representation;E2E-SLU models;subgroup identification;model bias analysis;divergence},
  doi={10.1109/TASLP.2024.3363447}}