from reazonspeech.k2.asr import load_model, transcribe, audio_from_path
import time
audio = audio_from_path("ja_test.wav")
model = load_model("cuda", "int8", "ja")
start_time = time.monotonic()
ret = transcribe(model, audio)
print(ret.text)
print(f"Decoding took {time.monotonic() - start_time:.2f} seconds")```
在4090和5090上都试过,结果是一样的。1个6s的音频文件,CPU耗时0.23s, GPU耗时0.66s
使用的示例进行测试: