app.py

from transformers import pipeline
import gradio as gr
import numpy as np
import librosa

# Initialize the speech recognition pipeline
pipe = pipeline("automatic-speech-recognition", model="oyemade/w2v-bert-2.0-yoruba-CV17.0")

def transcribe(audio):
    if audio is None:
        return "No audio detected. Please try again."
    
    try:
        # Check if the input is a file path (for uploaded files) or numpy array (for microphone input)
        if isinstance(audio, str):
            # Load the audio file using librosa
            audio, sr = librosa.load(audio, sr=16000)  # Resample to 16kHz
        elif isinstance(audio, tuple):  # Gradio audio components return a tuple (sr, audio)
            sr, audio = audio
            if sr != 16000:
                audio = librosa.resample(audio, sr, 16000)
        else:
            return "Invalid audio format. Please try again."

        # Check if the audio is valid (not silent)
        if np.max(np.abs(audio)) < 0.01:
            return "Audio is too quiet. Please speak louder or choose a different file and try again."
        
        text = pipe(audio)["text"]
        return text
    except Exception as e:
        return f"An error occurred: {str(e)}"

# Create the Gradio interface
iface = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.Audio(sources=["microphone", "upload"], type="filepath", label="Audio Input"),
    ],
    outputs="text",
    title="Neoform AI: Yoruba Speech Recognition",
    description="Realtime demo for Yoruba speech recognition using a fine-tuned Wav2Vec-Bert model. "
                "You can either use your microphone or upload an MP3 file. "
                "https://neoformai.com",
)

# Launch the interface
iface.launch()