Run SenseVoice model locally with sherpa-onnx

sensevoice_sherpa_onnx.py

import sherpa_onnx
import soundfile as sf
import numpy as np
from modelscope import snapshot_download

sensevoice_model_id = "WEAAEW/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17"
model_dir = snapshot_download(sensevoice_model_id, local_dir="./sensevoice")

recognizer = sherpa_onnx.OfflineRecognizer.from_sense_voice(
    model=f"{model_dir}/model.onnx",
    tokens=f"{model_dir}/tokens.txt",
    use_itn=True,
    debug=False,
)

def load_audio(filename: str):  
    """加载音频文件并返回 samples 和采样率"""  
    data, sample_rate = sf.read(  
        filename,  
        always_2d=True,  
        dtype="float32",  
    )
    data = data[:, 0]  # 使用第一个声道  
    samples = np.ascontiguousarray(data)  
    return sample_rate, samples

sample_rate, samples = load_audio("./Recording.wav")
stream = recognizer.create_stream()
stream.accept_waveform(sample_rate, samples)
recognizer.decode_stream(stream)
result = stream.result
print(result.text)