app.py

from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
import librosa
import torch
import gradio as gr

# Load Whisper model and processor
print("Loading model...")
processor = AutoProcessor.from_pretrained("jsbeaudry/whisper-medium-oswald")
model = AutoModelForSpeechSeq2Seq.from_pretrained("jsbeaudry/whisper-medium-oswald")
model.eval()

# Set device (GPU if available, else CPU)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
print("Model loaded successfully.")

# Transcription function
def transcribe(audio):
    if audio is None:
        return "Please upload or record an audio file first."

    # Gradio provides a tuple (sr, data)
    sr, data = audio

    # If stereo, convert to mono
    if len(data.shape) == 2:
        data = librosa.to_mono(data.T)

    # Resample to 16kHz if needed
    if sr != 16000:
        data = librosa.resample(data, orig_sr=sr, target_sr=16000)
        sr = 16000

    # Process audio
    input_features = processor(data, sampling_rate=sr, return_tensors="pt").input_features.to(device)

    # Predict
    with torch.no_grad():
        predicted_ids = model.generate(input_features)

    # Decode
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
    return transcription

# Gradio UI
def create_interface():
    with gr.Blocks(title="Whisper Medium - Haitian Creole") as demo:
        gr.Markdown("# 🎙️ Whisper Medium Creole ASR")
        gr.Markdown(
            "Upload or record your voice in Haitian Creole. Then click **Transcribe** to get the text."
        )

        with gr.Row():
            audio_input = gr.Audio(label="🎧 Upload or Record Audio", type="numpy", format="wav")
            transcribe_button = gr.Button("🔍 Transcribe")
            output_text = gr.Textbox(label="📝 Transcribed Text", lines=4)

        transcribe_button.click(fn=transcribe, inputs=audio_input, outputs=output_text)

    return demo

if __name__ == "__main__":
    interface = create_interface()
    interface.launch()