Datasets:

Aynursusuz
/

noisy-speech-dataset

license: apache-2.0
task_categories:
  - audio-classification
language:
  - en
tags:
  - audio
  - speech
  - noise-detection
  - audio-classification
size_categories:
  - 1K<n<10K

Noisy Speech Dataset

Binary classification dataset for detecting noisy audio in speech.

Dataset Description

This dataset is derived from haydarkadioglu/speech-noise-dataset with remapped labels:

speech: Clean speech audio (originally clean_speech)
noisy: Noisy audio including both noisy speech and noise-only samples (originally noisy_speech + noise_only)

Dataset Statistics

Split	Samples
Train	2873
Test	508

Label Distribution

speech: 1217 samples
noisy: 2164 samples

Usage

from datasets import load_dataset

# Load dataset
dataset = load_dataset("Aynursusuz/noisy-speech-dataset")

# Access train/test splits
train_data = dataset['train']
test_data = dataset['test']

# Example
print(train_data[0])

Model Training

from transformers import AutoModelForAudioClassification, TrainingArguments, Trainer

model = AutoModelForAudioClassification.from_pretrained(
    "MIT/ast-finetuned-audioset-10-10-0.4593",
    num_labels=2,
    label2id={"speech": 0, "noisy": 1},
    id2label={0: "speech", 1: "noisy"}
)

# Train your model...

Citation

Original dataset: haydarkadioglu/speech-noise-dataset

License

Apache 2.0