webui.py

import gradio as gr
import threading
import numpy as np
from scipy.io import wavfile
from scipy.signal import resample
from ali import *
from Zhipu import *


def to16000(sr, data, fname):
    # 计算重采样后的样本数量
    num_samples = round(len(data) * 16000 / sr)
    # 使用resample函数进行重采样
    resampled_data = resample(data, num_samples)
    # 创建新的1600Hz采样率的WAV文件
    wavfile.write(fname, 16000, resampled_data.astype(data.dtype))


# 这个函数将在后台运行，用于处理音频流并保存为WAV文件
def process_and_save_audio(sr, data, filename="recording.wav"):
    # 保存为WAV文件
    # write(filename, sr, data.astype(np.int16))
    to16000(sr, data, filename)


def audio2text(fname):
    accessKeyId = 'LTAI5tF5nA43VsaQApK6evWh'  # os.getenv('ALIYUN_AK_ID')
    accessKeySecret = 'ywESssmELzIorrIWGPvuL2pD9dbbgs'  # os.getenv('ALIYUN_AK_SECRET')
    appKey = 'lgo44Om6bL3j81AW'  # os.getenv('NLS_APP_KEY')
    # 执行录音文件识别
    url = upload2costemp(fname)
    resp = fileTrans(accessKeyId, accessKeySecret, appKey, url)
    if resp is None:
        return
    try:
        for i in range(len(resp['Result']['Sentences'])):
            s = resp['Result']['Sentences'][i]
            speakerid = s['SpeakerId']
            text = s['Text']
            texts.append(f'speaker_{speakerid}:{text}')
    except Exception as e:
        print(f'处理声音错误,{e}')

    # 初始化一个变量来跟踪上次保存文件的时间


last_save_time = time.time()
wavData = None
texts = []


def processWavData(sr, wavData):
    # 保存流到WAV文件
    fname = f"audio_{int(time.time())}.wav"
    process_and_save_audio(sr, wavData, fname)
    threading.Thread(target=audio2text, args=(fname,), daemon=True).start()


def transcribe(new_chunk):
    global last_save_time, wavData
    sr, y = new_chunk
    # print(f'change {y[:10]}')
    # print(f'{y[0:10]}, len(y)={len(y)}')

    if wavData is not None:
        wavData = np.concatenate((wavData, y))
    else:
        wavData = y
        last_save_time = time.time()

    # 检查是否已经过去了30秒
    if time.time() - last_save_time >= 30:
        processWavData(sr, wavData)
        # 重置流
        wavData = None
        # 更新上次保存时间
        last_save_time = time.time()
    return text_output  # transcriber({"sampling_rate": sr, "raw": stream})["text"]


def get_summary():
    _systemp = systemp.replace('{Data}', f'{",".join(texts)}')
    messages = [
        {
            "role": "system",
            "content": _systemp
        }
    ]
    print(_systemp)
    messages.append({"role": "user", "content": '请总结给出纪要'})
    response = client.chat.completions.create(
        model="glm-4-long",  # 填写需要调用的模型名称
        messages=messages,
        # tools=tools,
        tool_choice="auto",
    )
    print(response.choices[0].message)
    content = response.choices[0].message.content
    texts.append(f'总结:{content}')
    return texts  # summary(texts)


def startrecoding(new_chunk):
    global last_save_time
    # 更新上次保存时间
    last_save_time = time.time()


def stoprecoding(new_chunk):
    global wavData
    sr, y = new_chunk
    processWavData(sr, wavData)
    # 重置流
    wavData = None
    return texts


with gr.Blocks() as demo:
    gr.Label('Demo电子白板, 智能会议纪要生成工具')
    audio_input = gr.Audio(label="开始录音",
                           interactive=True, streaming=True,
                           waveform_options=gr.WaveformOptions(
                               sample_rate=16000,
                           ))
    text_output = gr.Textbox(label="转换后的文本", lines=10, interactive=False, value=texts)
    summary_button = gr.Button("获取总结")

    # 定义一个实时处理音频的函数，这里使用了 trans 函数
    audio_input.change(transcribe, inputs=audio_input, outputs=text_output)
    audio_input.start_recording(startrecoding, inputs=audio_input)
    audio_input.stop_recording(stoprecoding, inputs=audio_input, outputs=text_output)

    # 当按钮被点击时，调用 get_summary 函数，并将结果更新到 text_output 组件中
    summary_button.click(get_summary, outputs=text_output)

demo.launch(server_name="0.0.0.0", ssl_verify=False, server_port=8030)