app.py

import gradio as gr
import torch
import soundfile as sf
from transformers import AutoProcessor
from transformers.models.qwen2.modeling_qwen2 import Qwen2ForCausalLM

# 模型名称
MODEL_NAME = "Qwen/Qwen2-Audio-7B"

# 选择设备
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

# 加载处理器和模型
processor = AutoProcessor.from_pretrained(MODEL_NAME)
model = Qwen2ForCausalLM.from_pretrained(MODEL_NAME)
model.to(DEVICE)
model.eval()

# 转文字函数
def transcribe(audio_file):
    try:
        # 读取音频
        speech, rate = sf.read(audio_file.name)
        # 处理输入 - Qwen2-Audio使用audios参数
        inputs = processor(audios=speech, sampling_rate=rate, return_tensors="pt").to(DEVICE)
        # 推理生成
        generated_ids = model.generate(**inputs, max_length=512)
        # 解码
        transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)
        return transcription[0]
    except Exception as e:
        return f"出错了: {e}"

# Gradio 界面
with gr.Blocks() as demo:
    gr.Markdown("## Qwen2-Audio 7B 音频转文字 Demo")
    with gr.Row():
        audio_input = gr.Audio(source="upload", type="file", label="上传音频文件")
        output_text = gr.Textbox(label="识别结果")
    transcribe_btn = gr.Button("开始转文字")
    transcribe_btn.click(transcribe, inputs=audio_input, outputs=output_text)

# 启动
demo.launch()