app.py

#import gradio as gr
import tempfile
from pydub import AudioSegment
from transformers import pipeline
from pyannote.audio import Pipeline

# Load models dynamically
def load_models(model_size):
    if model_size == "transcriber":
        model_name = "clinifyemr/yoruba-model-finetuned"
        transcriber = pipeline("automatic-speech-recognition", model=model_name)
        return transcriber
    else:
        raise ValueError("Model size not supported in this application.")

# Process the audio file
def process_audio(file, num_speakers, model_size):
    audio_file = AudioSegment.from_file(file.name)
    transcriber = load_models(model_size)

    # Temporary file setup
    temp_path = tempfile.mktemp(suffix=".wav")
    audio_file.export(temp_path, format="wav")

    # Load diarization pipeline
    diarization_pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1", use_auth_token="HF_TOKEN")
    diarization = diarization_pipeline(temp_path, min_speakers=num_speakers, max_speakers=5)

    # Transcribe each segment
    def transcribe_segment(start, end):
        segment_audio = audio_file[start * 1000:end * 1000]  # pydub works in milliseconds
        segment_path = tempfile.mktemp(suffix=".wav")
        segment_audio.export(segment_path, format="wav")
        transcription = transcriber(segment_path)
        os.remove(segment_path)
        return transcription['text']

    transcripts = []
    for segment, _, speaker in diarization.itertracks(yield_label=True):
        transcription_text = transcribe_segment(segment.start, segment.end)
        transcripts.append(f"Speaker {speaker}: {transcription_text}")

    os.remove(temp_path)  # Clean up the temporary file
    return "\n".join(transcripts)

# Gradio interface setup
iface = gr.Interface(
    fn=process_audio,
    inputs=[
        #gr.components.Audio(label="Upload your audio file", type="file"),
        gr.components.Audio(label="Upload your audio file"),
        gr.components.Dropdown(choices=[1,2,3,4], label="Number of Speakers"),
        gr.components.Dropdown(choices=['transcriber'], label="Model Selection")  # Assuming only 'transcriber' is relevant here
    ],
    outputs=gr.Textbox(label="Transcription"),
    title="Audio Transcription and Speaker Diarization",
    description="Upload your audio file to transcribe and analyze speaker diarization."
)

if __name__ == "__main__":
    iface.launch()