| import gradio as gr | |
| from transformers import WhisperForConditionalGeneration, AutoProcessor | |
| model_name = "GiorgiSekhniashvili/whisper-tiny-ka-01" | |
| processor = AutoProcessor.from_pretrained(model_name) | |
| model = WhisperForConditionalGeneration.from_pretrained(model_name) | |
| def predict(audio): | |
| sr, waveform = audio | |
| input_values = processor(waveform, sampling_rate=16_000, return_tensors="pt") | |
| res = model.generate( | |
| input_values["input_features"], | |
| forced_decoder_ids=forced_decoder_ids, | |
| max_new_tokens=448, | |
| ) | |
| return processor.batch_decode(res, skip_special_tokens=True) | |
| mic = gr.Audio(source="microphone", type="numpy", label="Speak here...") | |
| demo = gr.Interface(predict, mic, "audio") | |
| if __name__ == "__main__": | |
| demo.launch() | |