stt_module.py

from faster_whisper import WhisperModel
import os

class STTManager:
    def __init__(self, model_size="base", device="cpu", compute_type="int8"):
        """
        model_size: tiny, base, small, medium, large-v3
        device: cpu or cuda
        compute_type: int8, float16, int8_float16 等
        """
        self.model_size = model_size
        self.device = device
        self.compute_type = compute_type
        self.model = None

    def load_model(self):
        """延迟加载模型，方便在加载前显示提示"""
        if self.model is None:
            # 自动处理 compute_type，如果 GPU 不支持 float16 则回退
            stt_compute_type = self.compute_type
            if self.device == "cpu" and stt_compute_type == "float16":
                stt_compute_type = "int8"
                
            self.model = WhisperModel(
                self.model_size, 
                device=self.device, 
                compute_type=stt_compute_type,
                download_root=os.path.join(os.getcwd(), "models")
            )
        return self.model

    def transcribe(self, audio_path):
        """识别音频并返回 segments (生成器) 和 info"""
        model = self.load_model()
        segments, info = model.transcribe(audio_path, beam_size=5)
        return segments, info

    @staticmethod
    def is_cuda_available():
        """检测 CUDA 是否可用"""
        try:
            import ctranslate2
            return ctranslate2.get_cuda_device_count() > 0
        except:
            return False

    @staticmethod
    def get_downloaded_models():
        """获取本地已下载的模型列表"""
        model_dir = os.path.join(os.getcwd(), "models")
        if not os.path.exists(model_dir):
            return []
        
        models = []
        for d in os.listdir(model_dir):
            if "faster-whisper-" in d:
                name = d.split("-")[-1]
                models.append(name)
            elif os.path.isdir(os.path.join(model_dir, d)):
                models.append(d)
        
        valid_names = ["tiny", "base", "small", "medium", "large-v1", "large-v2", "large-v3"]
        found = [m for m in models if any(v in m for v in valid_names)]
        final_models = []
        for f in found:
            for v in valid_names:
                if v in f:
                    final_models.append(v)
                    break
        
        return sorted(list(set(final_models)))