from transformers import AutoModelForAudioClassification, AutoFeatureExtractor
import torch
import torchaudio

# Load model and processor
model_checkpoint = "LincolnD/wav2vec2-base-finetuned-ravdess-personalization"
processor = AutoFeatureExtractor.from_pretrained(model_checkpoint)
model = AutoModelForAudioClassification.from_pretrained(model_checkpoint)

# Load and process audio
audio_path = "path/to/your/audio.wav"
waveform, sample_rate = torchaudio.load(audio_path)

# Resample to 16kHz if needed
if sample_rate != 16000:
    resampler = torchaudio.transforms.Resample(sample_rate, 16000)
    waveform = resampler(waveform)

# Process and predict
inputs = processor(waveform.squeeze().numpy(), sampling_rate=16000, return_tensors="pt", padding=True)
with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.argmax(outputs.logits, dim=-1)

emotion_id = predictions.item()
emotion = model.config.id2label[emotion_id]
print(f"Predicted emotion: {emotion}")

Training Details

This model was fine-tuned from the pre-trained Wav2Vec2 base model on the RAVDESS dataset.

Training Data

Dataset: RAVDESS
Emotions: 8 classes
Sampling Rate: 16kHz

Intended Use

This model is designed for use in Test-Time Adaptation (TTA) experiments as part of our research on adapting speech emotion recognition systems to new domains and speakers.

Evaluation

For detailed evaluation results and comparison with various TTA methods, please refer to our paper.

Model Card Authors

LincolnD

License

MIT License - See repository for details.