iqbalc/openAI-base

1import os2os.system("pip install git+https://github.com/openai/whisper.git")3 4import gradio as gr5import whisper6model = whisper.load_model("base")7 8import time9 10def transcribe(audio):11    # load audio for 30 seconds12    audio = whisper.load_audio(audio)13    audio = whisper.pad_or_trim(audio)14 15    # make log-Mel spectrogram and move to device as the model16    mel = whisper.log_mel_spectrogram(audio).to(model.device)17 18    # detect the spoken language19    _, probs = model.detect_language(mel)20    print(f"Detected language: {max(probs, key=probs.get)}")21 22    # decoding the audio23    options = whisper.DecodingOptions(fp16 = False)24    result = whisper.decode(model, mel, options)25    print(result.text)26    return result.text27    28gr.Interface(29    title = 'Speech to Text with OpenAI (base)', 30    fn=transcribe, 31    inputs=[32        gr.inputs.Audio(source="microphone", type="filepath")33    ],34    outputs=[35        "textbox"36    ],37    live=True).launch()