MedASR on Ascend NPU

pip install transformers==5.0.0 torch librosa soundfile
# torch-npu 需根据 CANN 版本安装，已预装

# 方式一：ModelScope
modelscope download --model google/medasr -d ./weights

# 方式二：AtomGit
python3 -m atomgit download hf_mirrors/google/medasr -d ./weights

python3 inference.py --model_path ./weights --audio ./weights/test_audio.wav --device npu

import torch
import librosa
from transformers import AutoModelForCTC, AutoProcessor

model_path = "./weights"
audio_path = "./weights/test_audio.wav"

device = "npu" if torch.npu.is_available() else "cpu"
processor = AutoProcessor.from_pretrained(model_path)
model = AutoModelForCTC.from_pretrained(model_path).to(device)
model.eval()

speech, sr = librosa.load(audio_path, sr=16000)
inputs = processor(speech, sampling_rate=sr, return_tensors="pt", padding=True)
inputs = {k: v.to(device) for k, v in inputs.items()}

with torch.no_grad():
    logits = model(**inputs).logits

predicted_ids = torch.argmax(logits, dim=-1)[0].tolist()

# CTC 解码：去空白、合并重复
prev = None
ctc_ids = []
for pid in predicted_ids:
    if pid != 0 and pid != prev:
        ctc_ids.append(pid)
    prev = pid

transcription = processor.tokenizer.decode(ctc_ids, skip_special_tokens=True)
print(transcription)

[EXAM TYPE] CT chest PE protocol {period} [INDICATION] 54-year-old female, shortness of breath, evaluate for PE {period} [TECHNIQUE] Standard protocol {period} [FINDINGS] {colon} Pulmonary vasculature {colon} The main PA is patent {period} There are filling defects in the segmental branches of the right lower lobe {comma} compatible with acute PE {period} No saddle embolus {period} Lungs {colon} No pneumothorax {period} Small bilateral effusions {comma} right greater than left {period} {new paragraph} [IMPRESSION] {colon} Acute segmental PE, right lower lobe {period}

python3 benchmark.py --model_path ./weights --audio ./weights/test_audio.wav --device npu --warmup 3 --iterations 10

python3 accuracy.py --model_path ./weights --audio ./weights/test_audio.wav

def apply_npu_monkey_patch():
    if hasattr(torch, "npu") and torch.npu.is_available():
        torch.cuda = torch.npu
        torch.cuda.is_available = torch.npu.is_available
        if hasattr(torch.npu, "current_device"):
            torch.cuda.current_device = torch.npu.current_device