FizzyBrain/ast-nonspeech7k-finetuned

AST Fine-tuned for Non-Speech Sound Classification

This model is a fine-tuned version of MIT/ast-finetuned-audioset-10-10-0.4593 on the Nonspeech7k dataset.

Model Details

• Base Model: MIT/ast-finetuned-audioset-10-10-0.4593
• Fine-tuned on: Nonspeech7k dataset
• Classes: breath, cough, crying, laugh, screaming, sneeze, yawn
• Sample Rate: 16kHz
• Input Length: 10 seconds (160,000 samples)

Usage

from transformers import ASTFeatureExtractor, ASTForAudioClassification
import torch
import torchaudio

# Load model
feature_extractor = ASTFeatureExtractor.from_pretrained("FizzyBrain/ast-nonspeech7k-finetuned")
model = ASTForAudioClassification.from_pretrained("FizzyBrain/ast-nonspeech7k-finetuned")

# Load and preprocess audio
waveform, sample_rate = torchaudio.load("audio.wav")
inputs = feature_extractor(waveform, sampling_rate=16000, return_tensors="pt")

# Predict
with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predicted_class_id = predictions.argmax().item()

Classes

0. breath
1. cough
2. crying
3. laugh
4. screaming
5. sneeze
6. yawn

Training Details

• Fine-tuned using advanced augmentation techniques
• Class-weighted loss for imbalanced data
• Layer-wise learning rate decay
• Early stopping with macro-F1 monitoring