Audio Speed-Level Classifier (Single-Label)

This model is a fine-tuned version of openai/whisper-small for single-label classification. It categorizes the speaking rate of an audio clip into one of three distinct speed levels.

🏷️ Speed Labels

• fast speed: Rapid speech, often with elided syllables.
• measured speed: Standard, professional, or moderate speaking pace.
• slow speed: Deliberate, calm, or hesitant speech.

---

🛠 Usage: Input & Output

1. Input Specifications

• Processor: Uses WhisperProcessor for resampling and Mel-spectrogram generation.
• Sampling Rate: 16,000 Hz (Standard Whisper requirement).
• Audio Format: Mono raw waveform.

2. Output (Single-Label Logic)

Because this is a Single-Label task, the categories are mutually exclusive.

• Activation: Softmax. This ensures all output probabilities sum exactly to 1.0 (100%).
• Decision: The model selects the label with the highest probability score.

---

🚀 Inference Code

import torch
import librosa
import numpy as np
from transformers import WhisperProcessor, WhisperForAudioClassification

model_id = "Kang-Chieh/whisper-small-speed_level"
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load the processor and model
processor = WhisperProcessor.from_pretrained(model_id)
model = WhisperForAudioClassification.from_pretrained(model_id).to(device)

def predict_speed(audio_path):
    # 1. Load audio and ensure 16kHz
    audio, _ = librosa.load(audio_path, sr=16000)
    
    # 2. Preprocess
    inputs = processor(audio, sampling_rate=16000, return_tensors="pt").to(device)

    # 3. Inference
    with torch.no_grad():
        logits = model(**inputs).logits
    
    # 4. Single-label logic (Softmax)
    probs = torch.softmax(logits, dim=-1).squeeze().cpu().numpy()
    
    # 5. Get the highest scoring label
    id2label = model.config.id2label
    predicted_id = np.argmax(probs)
    
    return {
        "label": id2label[predicted_id],
        "confidence": float(probs[predicted_id]),
        "all_scores": {id2label[i]: float(probs[i]) for i in range(len(probs))}
    }

# Run example
result = predict_speed("audio_clip.wav")
print(f"Detected Speed: {result['label']} ({result['confidence']:.2%})")