'''Librispeech 100h English ASR demo @ML2 2022-02-11 ''' import os from glob import glob from loguru import logger import soundfile as sf import gradio as gr from espnet_model_zoo.downloader import ModelDownloader from espnet2.bin.asr_inference import Speech2Text # ---------- Settings ---------- GPU_ID = '-1' os.environ['CUDA_VISIBLE_DEVICES'] = GPU_ID DEVICE = 'cuda' if GPU_ID != '-1' else 'cpu' SERVER_PORT = 42208 SERVER_NAME = "0.0.0.0" SSL_DIR = './keyble_ssl' MODEL_DIR = '/home/jkang/HDD4T/jkang/huggingface' EXAMPLE_DIR = './examples' examples = sorted(glob(os.path.join(EXAMPLE_DIR, '*.wav'))) # ---------- Logging ---------- logger.add('app.log', mode='a') logger.info('============================= App restarted =============================') # ---------- Model ---------- logger.info('download model') d = ModelDownloader(MODEL_DIR) out = d.download_and_unpack("jkang/espnet2_librispeech_100_conformer") logger.info('model downloaded') model = Speech2Text.from_pretrained( asr_train_config=out['asr_train_config'], asr_model_file=out['asr_model_file'] ) logger.info('model loaded') def predict(wav_file): logger.info('wav file loaded') speech, rate = sf.read(wav_file) nbests = model(speech) text, *_ = nbests[0] logger.info('predicted') return text iface = gr.Interface( predict, title='영어 음성인식 데모 (espnet libri100) -- 프로토타입', description='영어 음성 파일을 업로드하면 텍스트 내용을 결과로 보여줍니다.', inputs=[ gr.inputs.Audio(label='영어 음성', source='upload', type='filepath') ], outputs=[ gr.outputs.Textbox(label='음성 인식 디코딩결과'), ], examples=examples, article='

i-Scream AI

', ) if __name__ == '__main__': try: iface.launch(debug=True, server_name=SERVER_NAME, server_port=SERVER_PORT, enable_queue=True, # ssl_keyfile=SSL_DIR, # ssl_certfile=SSL_DIR ) except KeyboardInterrupt as e: print(e) finally: iface.close()