app.py

import gradio as gr
import whisperx



def transcribe_audio(whisper_model,filename,whisper_compute_type,whisper_device="cpu",whisper_batch_size=4,source_lang=None):
    # model = whisperx.load_model(whisper_model, device="cuda", compute_type="float16")
    # global WHISPER_MODEL
    
    WHISPER_MODEL = whisperx.load_model(whisper_model,device=whisper_device,compute_type=whisper_compute_type)
    
    print("Whisper model loaded")
    
    audio = whisperx.load_audio(filename)
    
    language_for_whisper = None if source_lang == "auto" else source_lang
    
    result = WHISPER_MODEL.transcribe(audio, batch_size=whisper_batch_size,language=language_for_whisper)
    language = result["language"]
    segments = result["segments"]
    
    # print(segments)

    print(f"Detected language {language}")
    

    if source_lang == "auto":
        detected_language = language  
    else:
        detected_language = source_lang
        
    # Formated text
    ready_text = ""
    
    for segment in segments:
        ready_text+=segment["text"] + "\n"

    return ready_text

def process_audio(audio_input,whisper_model,whisper_language,whisper_compute_type):
    result= transcribe_audio(filename=audio_input, whisper_model=whisper_model,source_lang=whisper_language,whisper_compute_type=whisper_compute_type)
    return result

with gr.Blocks() as demo:
    gr.Markdown("# Any whisper ct2 test")

    with gr.Row():
        with gr.Column(scale=1):
            with gr.Accordion(open=False,label="Whisper settings"):
                    whisper_model = gr.Dropdown(choices=["tiny","daswer123/whisper-medium-uz-ct2"],value="daswer123/whisper-medium-uz-ct2", allow_custom_value=True)
                    whisper_language = gr.Dropdown(choices=["auto","uz"],value="uz", allow_custom_value=True)
                    whisper_compute_type = gr.Dropdown(choices=["int8","float32"],value="int8")
            audio_input = gr.Audio(label="Upload your audio",type="filepath")

        with gr.Column(scale=1):
            text_output = gr.Textbox(label="Result")
            submit_btn = gr.Button("Start!")

    submit_btn.click(fn=process_audio, inputs=[audio_input,whisper_model,whisper_language,whisper_compute_type], outputs=text_output)

demo.launch()
# result = transcribe_audio("tiny","bezbeka.mp3","float32",source_lang="uz")
# print(result)