shuyuncci/wavlm-accent

WavLM Large — English Accent Classification

This model is a fine-tuned version of microsoft/wavlm-large for classifying English accents from audio. Only the classification head was trained; the WavLM backbone weights are frozen.

Model Details

• Base model: microsoft/wavlm-large
• Task: Audio Classification
• Accents: 8 classes
• Training: Classification head only (backbone frozen)
• Framework: HuggingFace Transformers

Supported Accents

ID	Label
0	american
1	australian
2	british
3	canadian
4	indian
5	irish
6	jamaican
7	scottish

How to Use

Basic Inference

import torch
import torchaudio
from transformers import AutoFeatureExtractor, AutoModelForAudioClassification

# Load model and feature extractor
model_id = "shuyuncci/wavlm-accent"
feature_extractor = AutoFeatureExtractor.from_pretrained(model_id)
model = AutoModelForAudioClassification.from_pretrained(model_id)
model.eval()

# Load audio file (must be 16kHz mono)
waveform, sample_rate = torchaudio.load("your_audio.wav")

# Resample if needed
if sample_rate != 16000:
    resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
    waveform = resampler(waveform)

# Convert to mono if stereo
if waveform.shape[0] > 1:
    waveform = waveform.mean(dim=0, keepdim=True)

# Prepare input
inputs = feature_extractor(
    waveform.squeeze().numpy(),
    sampling_rate=16000,
    return_tensors="pt"
)

# Run inference
with torch.no_grad():
    logits = model(**inputs).logits

# Get predicted label
predicted_id = logits.argmax(dim=-1).item()
predicted_label = model.config.id2label[predicted_id]
print(f"Predicted accent: {predicted_label}")

Important Notes

• Input format: 16kHz mono WAV audio
• Audio length: The model was trained on clips up to 20 seconds. For longer audio, consider splitting into chunks and aggregating predictions.
• Language: This model is designed for English speech only. Performance on non-English audio is not guaranteed.