MeUtils 2024.3.4.13.4.45__py3-none-any.whl → 2025.1.16.17.15.52__py3-none-any.whl

meutils/ai_audio/asr/openai_asr.py

@@ -6,16 +6,93 @@
 # @Author       : betterme
 # @WeChat       : meutils
 # @Software     : PyCharm
-# @Description  : 
+# @Description
+import io
+import os
+
+import requests
+from meutils.io.files_utils import to_file
 
 from meutils.pipe import *
 from openai import OpenAI
 
+# response_format: Literal["json", "text", "srt", "verbose_json", "vtt"]
+file = open(
+    "f275d9ac-5a62-4bbe-baf9-3fa10e0332f4.mp3",
+    'rb')  # 正确
+# file=open("2022112519张健涛29.mp3", 'rb')
+# file = httpx.get("https://oss.chatfire.cn/data/demo.mp3").content
+
 client = OpenAI(
-    api_key=os.getenv('OPENAI_API_KEY'),
-    base_url=os.getenv('OPENAI_API_BASE'),
-    max_retries=3
+    # base_url="https://openai.chatfire.cn/v1",
+    api_key=os.getenv("OPENAI_API_KEY") + "-336",
+    # api_key=os.getenv("OPENAI_API_KEY_OPENAI") + "-336",
+
+    # api_key=os.getenv("SILICONFLOW_API_KEY"),
+    # base_url=os.getenv("SILICONFLOW_BASE_URL")
+
+    # base_url="http://0.0.0.0:8000/v1",
+    # base_url="https://api.chatfire.cc/v2",
+    # base_url="https://api.chatfire.cc/v1",
+
+    # api_key=os.getenv("DEEPINFRA_API_KEY"),
+    # base_url="https://api.deepinfra.com/v1/openai",
 )
+# client = OpenAI(
+#     api_key=os.environ.get("GROQ_API_KEY"),
+#     # base_url=os.getenv("GROQ_BASE_URL"),
+#     base_url="https://groq.chatfire.cc/v1",
+#
+# )
+with timer():
+    file = Path('audio.mp3').read_bytes()
+    file = Path("/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/x1.mp3").read_bytes()
+
+    # file = Path('m4a.mp3')
+
+    # file = Path("/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/example.mp3").read_bytes()
+
+    # file = Path("/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/国外健身和生活设备市场观察及分析.mp3").read_bytes()
+
+    # file = Path('audio.mp3').read_bytes()
+
+    # file = arun(to_file(file))  # mp3后缀
+    file = ('_.mp3', file)  # ok
+
+    _ = client.audio.transcriptions.create(
+        file=file,
+
+        # model="whisper-1",
+        # model="openai/whisper-large",
+
+        # model="SenseVoiceSmall",
+        model="FunAudioLLM/SenseVoiceSmall",
+
+        # model="distil-whisper/distil-large-v3-ct2",
+
+        # model="asr",
+
+        # model="sensevoice",
+        # model="whisper-large-v3",
+        # response_format="text",  # ["json", "text", "srt", "verbose_json", "vtt"]
+        # response_format="srt",  # ["json", "text", "srt", "verbose_json", "vtt"]
+        # response_format="verbose_json",  # ["json", "text", "srt", "verbose_json", "vtt"]
+        # response_format="vtt",  # ["json", "text", "srt", "verbose_json", "vtt"]
+
+    )
+    print(_)
+    #
+    # _ = client.audio.translations.create(
+    #     file=file,
+    #     # model="whisper-1",
+    #     model="whisper-large-v3",
+    #     # response_format="text",  # ["json", "text", "srt", "verbose_json", "vtt"]
+    #     # response_format="srt",  # ["json", "text", "srt", "verbose_json", "vtt"]
+    #     # response_format="verbose_json",  # ["json", "text", "srt", "verbose_json", "vtt"]
+    #     # response_format="vtt",  # ["json", "text", "srt", "verbose_json", "vtt"]
+    #
+    # )
+    # print(_)
+
+# Transcription(text='健身需要注意适度和平衡 过度的锻炼可能会导致身体受伤 因此 进行健身活动前 最好先咨询医生 或专业的健身教练 制定一个适合自己的健身计划 一般来说 一周内进行150分钟的适度强度的有氧运动 或者75分钟的高强度有氧运动 加上每周两天的肌肉锻炼就能达到保持健康的目标')
 
-_ = client.audio.transcriptions.create(file=open("hi.wav", 'rb'), model="whisper-1")
-print(_)

meutils/ai_audio/fast_asr.py

@@ -6,8 +6,8 @@
 # @Author       : betterme
 # @WeChat       : meutils
 # @Software     : PyCharm
-# @Description  : 
-import pysrt
+# @Description  :
+
 from datetime import timedelta
 
 from meutils.pipe import *
@@ -51,8 +51,9 @@ class ASR(object):
             yield segment
 
     def to_srt(self, segments: Generator[Segment, Any, None], filename='subtitle.srt'):
-        subs = pysrt.SubRipFile()
+        import pysrt
 
+        subs = pysrt.SubRipFile()
         for index, segment in enumerate(segments):  # todo: index 从0开始？
             subs.append(
                 pysrt.SubRipItem(
@@ -82,6 +83,8 @@ class ASR(object):
         time_interval = timedelta(seconds=seconds)
 
         # 创建一个 SubRipTime 对象
+        import pysrt
+
         srt_time = pysrt.SubRipTime(
             hours=time_interval.seconds // 3600,
             minutes=(time_interval.seconds // 60) % 60,
@@ -95,6 +98,7 @@ class ASR(object):
 if __name__ == '__main__':
     # audio_path = '../../../zh_.wav'
     audio_path = "/Users/betterme/Downloads/videoplayback.m4a"
-    asr = ASR('large')
+    # asr = ASR('large')
+    asr = ASR('tiny', local_files_only=False)
 
     asr.audio_to_srt(audio_path)

meutils/ai_audio/tts/EdgeTTS.py

@@ -7,6 +7,7 @@
 # @WeChat       : meutils
 # @Software     : PyCharm
 # @Description  : 增加固定时长的生成
+# https://github.com/rany2/edge-tts/blob/master/examples/streaming_with_subtitles.py
 
 import edge_tts
 
@@ -48,7 +49,7 @@ class EdgeTTS(object):
             volume: float = 0,
             filename: Optional[None] = None,
     ):
-        voices = (await edge_tts.VoicesManager.create()).find(**{'ShortName': role}) or [{}]
+        voices = (await edge_tts.VoicesManager.create()).find(**{'ShortName': role}) or [{}]  # todo: 优化
         voice = voices[0].get('ShortName', 'zh-CN-YunxiNeural')
 
         rate = f"{'+' if rate >= 0 else ''}{rate}%"
@@ -58,13 +59,14 @@ class EdgeTTS(object):
 
         filename = filename or f"{time.time()}.mp3"  # f"{str(datetime.datetime.now())[:19]}.mp3"
         await communicate.save(filename)
+
         return filename
 
-    @retrying
+    @retrying()
     async def stream_acreate(
             self,
             text: str,
-            voice: str = '云希',  # "晓晓", "晓伊", "云健", "云希", "云夏", "云扬", "晓北辽宁", "陕西晓妮"
+            voice: str = '云希',  # "晓晓", "晓伊", "云健", "云希", "云夏", "云扬", "辽宁晓北", "陕西晓妮"
             rate: float = 0,
             volume: float = 0,
             **kwargs
@@ -87,10 +89,10 @@ class EdgeTTS(object):
         communicate = edge_tts.Communicate(text, voice=voice, rate=rate, volume=volume)
         async for chunk in communicate.stream():  # 构建生成器，流式输出， todo: 字幕结构体
             # {'type': 'WordBoundary', 'offset': 1000000, 'duration': 6750000, 'text': '你好'} # 断句字幕
+            # https://github.com/rany2/edge-tts/blob/master/examples/streaming_with_subtitles.py
             # if chunk.get('type') == 'WordBoundary':
             #     logger.debug(chunk)
             #     continue
-
             if chunk.get('type') == "audio":
                 yield chunk["data"]
 
@@ -103,6 +105,19 @@ class EdgeTTS(object):
         """
         return asyncio.run(edge_tts.VoicesManager.create()).find(**kwargs or {"Locale": "zh-CN"})
 
+    async def acreate_for_openai(
+            self,
+            input: str,
+            model: str = 'edge-tts',
+            voice: str = '云希',  # 男声 女声
+            **kwargs
+    ):
+        data = {
+            "text": input,
+            "voice": voice,
+        }
+        return self.stream_acreate(**data)
+
 
 if __name__ == '__main__':
     # print(EdgeTTS().create(['不知道'] * 10))
@@ -119,6 +134,17 @@ if __name__ == '__main__':
     陕西省，简称“陕”或“秦”，中华人民共和国省级行政区，省会西安，位于中国内陆腹地，黄河中游，东邻山西、河南，西连宁夏、甘肃，南抵四川、重庆、湖北，北接内蒙古，介于东经105°29′—111°15′，北纬31°42′—39°35′之间，总面积205624.3平方千米。 [1] [5]截至2022年11月，陕西省下辖10个地级市（其中省会西安为副省级市）、31个市辖区、7个县级市、69个县。 [121]截至2022年末，陕西省常住人口3956万人。
     """
 
-    # print(cls.find_voices(Locale="zh-CN"))
-    for i in async2sync_generator(cls.stream_acreate(input)):
-        print(i)
+    print(cls.find_voices(Locale="zh-CN"))
+    # for i in async2sync_generator(cls.stream_acreate(input)):
+        # print(i)
+
+    # EDGE_TTS_DICT = {
+    #     "用英语": "en-US-AriaNeural",
+    #     "用日语": "ja-JP-NanamiNeural",
+    #     "用法语": "fr-BE-CharlineNeural",
+    #     "用韩语": "ko-KR-SunHiNeural",
+    #     "用德语": "de-AT-JonasNeural",
+    #     # add more here
+    # }
+
+    arun(cls.acreate(input))

meutils/ai_audio/tts/openai_tts.py

@@ -6,17 +6,22 @@
 # @Author       : betterme
 # @WeChat       : meutils
 # @Software     : PyCharm
-# @Description  :
+# @Description  : todo: 更多模型
 
 from meutils.pipe import *
 from openai import OpenAI
 
-base_url = os.getenv('OPENAI_API_BASE')
-base_url = 'http://0.0.0.0:8000/v1'
+# base_url = 'http://0.0.0.0:8000/v1'
+# base_url='http://154.3.0.117:39009/tts/v1'
+# base_url="https://openai-dev.chatfire.cn/audio/v1"
+# base_url="https://openai.chatfire.cn/audio/v1"
+# base_url="https://api.chatfire.cn/v2"
+
 client = OpenAI(
-    api_key=os.getenv('OPENAI_API_KEY'),
-    base_url=base_url,
-    max_retries=3
+    # base_url=base_url,
+    # api_key=os.getenv('OPENAI_API_KEY'),
+
+    # api_key=os.getenv('OPENAI_API_KEY_OPENAI'),
 )
 
 # text = """
@@ -26,17 +31,16 @@ client = OpenAI(
 # _ = client.audio.speech.create(input=text, model="tts", voice="晓晓")
 # _.stream_to_file("hi.mp3")
 
-
-# from pathlib import Path
-from openai import OpenAI
-
-client = OpenAI(api_key='sk-TZVGYMoCUYrGSddb56F167A54e544e658aD5F1B8D809CfF4', base_url=base_url)
-
-speech_file_path = "speech_xx.mp3"
-response = client.audio.speech.create(
-    input="健身需要注意适度和平衡，过度的锻炼可能会导致身体受伤。因此，进行健身活动前，最好先咨询医生或专业的健身教练，制定一个适合自己的健身计划。一般来说，一周内进行150分钟的适度强度的有氧运动，或者75分钟的高强度有氧运动，加上每周两天的肌肉锻炼，就能达到保持健康的目标。",
-    model="tts",
-    voice="晓晓",
-)
-
-response.stream_to_file(speech_file_path)
+with timer():
+    voice = "晓晓"
+    speech_file_path = f"{voice}.mp3"
+    response = client.audio.speech.create(
+        # input="健身需要注意适度和平衡，过度的锻炼可能会导致身体受伤。因此，进行健身活动前，最好先咨询医生或专业的健身教练，制定一个适合自己的健身计划。一般来说，一周内进行150分钟的适度强度的有氧运动，或者75分钟的高强度有氧运动，加上每周两天的肌肉锻炼，就能达到保持健康的目标。",
+        input="健身需要注意适度和平衡，过度的锻炼可能会导致身体受伤",
+
+        model="tts-1",
+        voice=voice,
+        # voice="54a5170264694bfc8e9ad98df7bd89c3"
+    )
+    # logger.debug(response)
+    response.stream_to_file(speech_file_path)

meutils/ai_audio/tts/tts_ui.py

@@ -9,6 +9,7 @@
 # @Description  : https://www.qianduan.shop/blogs/detail/140
 # https://baijiahao.baidu.com/s?id=1762431177147996323&wfr=spider&for=pc
 # https://github.com/pndurette/gTTS
+# https://github.com/ycyy/edge-tts-webui
 
 from meutils.pipe import *
 import gradio as gr

meutils/ai_audio/utils.py

@@ -42,3 +42,12 @@ def merge_audios(audio_files, filename: Optional[str] = None):
 
 def to_audio(file, filename="example.wav", format='wav', **kwargs):
     AudioSegment.from_file(file).export(filename, format=format, **kwargs)
+
+
+if __name__ == '__main__':
+    file = "/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/asr/shaoyinceo.m4a"
+    file = "/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/asr/audio-.mp3"
+    file = "/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/x1.wav"
+
+    to_audio(file, filename="x1.mp3", format='mp3')
+

meutils/ai_cv/__init__.py

@@ -8,4 +8,3 @@
 # @Software     : PyCharm
 # @Description  : 
 
-from meutils.pipe import *

meutils/ai_cv/ocr.py

@@ -17,9 +17,10 @@ from rapidocr_onnxruntime import RapidOCR
 
 rapid_ocr = RapidOCR()
 # results, elapse = rapid_ocr('invoice.jpg')
-results, elapse = rapid_ocr('tbl.png')
+# results, elapse = rapid_ocr('tbl.png')
+results, elapse = rapid_ocr('bg.png')
 
 rprint(results)
 print([i[1] for i in results])
 
-# rprint([dict(zip(['坐标', '文字'], r)) for r in results])
+# rprint([dict(zip(['坐标', '文字'], r)) for r in results])

meutils/ai_cv/utils.py

@@ -0,0 +1,154 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# @Project      : AI.  @by PyCharm
+# @File         : utils
+# @Time         : 2024/6/12 08:34
+# @Author       : betterme
+# @WeChat       : meutils
+# @Software     : PyCharm
+# @Description  :
+
+from meutils.pipe import *
+from meutils.oss.minio_oss import Minio
+
+from PIL import Image, ImageDraw
+
+
+def crop_polygon(image: Union[str, bytes], outline_points, inline_points):
+    if isinstance(image, bytes):
+        image = io.BytesIO(image)
+
+    # 打开图像
+    img = Image.open(image)
+
+    # 创建一个与原图大小相同的黑色遮罩
+    mask = Image.new('L', img.size, 0)
+
+    # 在遮罩上绘制白色多边形
+    for points in outline_points:
+        ImageDraw.Draw(mask).polygon(points, outline=1, fill=255)
+
+    for points in inline_points:
+        ImageDraw.Draw(mask).polygon(points, outline=1, fill=0)
+
+    # 将遮罩应用到原图
+    output = Image.new('RGBA', img.size, (0, 0, 0, 0))
+    output.paste(img, (0, 0), mask)
+
+    # # 将 PIL Image 转换为字节
+    # buffer = io.BytesIO()
+    # output.save(buffer, format="PNG")
+    # byte_data = buffer.getvalue()
+    #
+    # return byte_data
+    return output
+
+
+def crop_image(
+        image: Union[Path, str, bytes],
+
+        width: Optional[int] = None,
+        height: Optional[int] = None,
+        x: int = 0,
+        y: int = 0,
+
+        debug: bool = False
+):
+    """
+        crop_image(image="https://sfile.chatglm.cn/testpath/4f520e7f-c2f5-5555-8d1f-4fda0fec0e9d_0.png", height=950, debug=True)
+
+    :param image:
+    :param width:
+    :param height:
+    :param x:
+    :param y:
+    :param debug:
+    :return:
+    """
+    if isinstance(image, bytes):
+        image = io.BytesIO(image)
+
+    elif isinstance(image, str) and image.startswith("http"):
+        response = httpx.get(image)
+        image = io.BytesIO(response.content)
+
+    image = Image.open(image)  # .convert('RGB')
+
+    width = width or image.width
+    height = height or image.height
+    cropped_image = image.crop((x, y, x + width, y + height))
+
+    if debug:
+        logger.debug(image.size)
+        logger.debug(cropped_image.size)
+        cropped_image.show()
+
+    return cropped_image  # .save('output.png')
+
+
+def crop_image_and_upload(
+        image: Union[Path, str, bytes],
+
+        width: Optional[int] = None,
+        height: Optional[int] = None,
+        x: int = 0,
+        y: int = 0,
+
+        debug: bool = False,
+
+        file_id: Optional[str] = None
+):
+    # 创建一个 BytesIO 对象用于保存字节数据
+    # byte_io = io.BytesIO()
+
+    # 将图像保存到 BytesIO 对象中
+    # image.save(byte_io, format='PNG')
+    # 获取字节数据
+    # byte_data = byte_io.getvalue()
+
+    image = crop_image(image, width, height, x, y, debug)
+
+    file_id = file_id or shortuuid.random()
+    filename = f"{file_id}.png"
+    image.save(filename)
+
+    # 上传
+    client = Minio()
+    client.fput_object('files', filename, file_path=filename, content_type='image/png')
+
+    # 删除临时文件
+    Path(filename).unlink(True)
+    return client.get_file_url(filename)
+
+
+def del_watermark():  # 去水印
+    import poimage
+    # poimage = try_import("poimage")
+    # 支持jpg、png等所有图片格式
+    poimage.del_watermark(
+        input_image="img.png",
+        output_image='img_.png'
+    )
+
+
+if __name__ == '__main__':
+    # (left, top, right, bottom)
+
+    # 调用该函数传入图片路径、裁剪起始点坐标(x, y)以及宽度和高度
+    # crop_image("yuanbao.jpeg", height= 998, debug=True)
+    # crop_image("zhipu.png", 0, 0, 1024, 950)
+    # crop_image(Path("yuanbao.jpeg").read_bytes(), height=998, debug=True)
+
+    # r = requests.get("https://sfile.chatglm.cn/testpath/4f520e7f-c2f5-5555-8d1f-4fda0fec0e9d_0.png")
+    # print(r.content)
+
+    image = crop_image(
+        image="https://sfile.chatglm.cn/testpath/4f520e7f-c2f5-5555-8d1f-4fda0fec0e9d_0.png",
+        height=950,
+        debug=True
+    )
+
+    print(crop_image_and_upload(
+        image="https://sfile.chatglm.cn/testpath/4f520e7f-c2f5-5555-8d1f-4fda0fec0e9d_0.png",
+        height=950,
+        debug=True))

meutils/ai_video/avmerge.py

@@ -33,3 +33,9 @@ video_clip.write_videofile("avmerged3.mp4")
 # FileNotFoundError: [Errno 2] No such file or directory:\
 #     'ffmpeg -hide_banner -y -i "/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_video/output.mp4" ' \
 #     '-ac 1 "/Users/betterme/PycharmProjects/AI/pyvideotrans/tmp/output/output.wav"'
+
+
+def extract_audio(video__path, audio_path):
+    my_clip = AudioFileClip(video__path)
+    my_clip.write_audiofile(audio_path)
+    return my_clip

meutils/ai_video/video.py

@@ -25,7 +25,7 @@ def video2image(video: str, top_duration=180):
             clip_.save_frame(p / f'{i}.png')
 
 
-def video2audio(path_pair, verbose=False, subclip=None, ffmpeg_params=["-f", "mp3"]):
+def _video2audio(path_pair, verbose=False, subclip=None, ffmpeg_params=["-f", "mp3"]):
     """
         clip = VideoFileClip('蛋清和蛋黄是这么分离的.720p').subclip(3, 7)
 
@@ -48,6 +48,15 @@ def video2audio(path_pair, verbose=False, subclip=None, ffmpeg_params=["-f", "mp
         )
 
 
+def video2audio(video_path, audio_path):
+    with AudioFileClip(video_path) as clip:
+        clip.write_audiofile(audio_path)
+
+
 if __name__ == '__main__':
+    video_path = "douyin.mp4"
+
     # video2image("/Users/betterme/Downloads/11月15日.mp4")
-    video2audio(["/Users/betterme/Downloads/11月15日.mp4", "x.mp3"])
+    # video2audio([video_path, "x.mp3"])
+
+    print(video2audio('/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/asr/blibli.mp4', "x.mp3"))

meutils/{api → apis}/__init__.py

@@ -8,4 +8,4 @@
 # @Software     : PyCharm
 # @Description  : 免费api https://www.cnblogs.com/Stars-are-shining/p/13383867.html
 
-from meutils.api.common import *
+from meutils.apis.common import *

meutils/apis/ali_apis.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# @Project      : AI.  @by PyCharm
+# @File         : express
+# @Time         : 2024/4/24 17:17
+# @Author       : betterme
+# @WeChat       : meutils
+# @Software     : PyCharm
+# @Description  :
+
+from meutils.pipe import *
+
+@lru_cache
+def express_query(express_no=780098068058, path: str = '/kdi', appcode: Optional[str] = None):
+    """
+
+    :param appcode:
+        appcode = '0ccd86184de94ca19c37cbb215b1f3722'  # 开通服务后 买家中心-查看AppCode
+
+    :param express_no:
+    :param path:
+    :return:
+    """
+    base_url = "https://wuliu.market.alicloudapi.com"
+    with httpx.Client(base_url=base_url) as client:
+        headers = {"Authorization": f'APPCODE {appcode or os.getenv("EXPRESS_APP_CODE")}'}
+        params = {'no': express_no}
+
+        response = client.get(path, params=params, headers=headers)
+        if response.status_code == 200:
+            return response.json()
+        else:
+            return {"error": response.text}
+
+
+async def aexpress_query(express_no=780098068058, path: str = '/kdi', appcode: Optional[str] = None):
+    """
+
+    :param appcode:
+        appcode = '0ccd86184de94ca19c37cbb215b1f372'  # 开通服务后 买家中心-查看AppCode
+
+    :param express_no:
+    :param path:
+    :return:
+    """
+    base_url = "https://wuliu.market.alicloudapi.com"
+    async with httpx.AsyncClient(base_url=base_url) as client:
+        headers = {"Authorization": f'APPCODE {appcode or os.getenv("EXPRESS_APP_CODE")}'}
+        params = {'no': express_no}
+        response = await client.get(path, params=params, headers=headers)
+        if response.status_code == 200:
+            return response.json()
+        else:
+            return {"error": response.text}
+
+
+if __name__ == '__main__':
+    # print(express_query('0ccd86184de94ca19c37cbb215b1f372'))
+
+    print(arun(aexpress_query('0ccd86184de94ca19c37cbb215b1f372')))

meutils/apis/audio/__init__.py

@@ -0,0 +1,10 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# @Project      : AI.  @by PyCharm
+# @File         : __init__
+# @Time         : 2024/7/18 19:16
+# @Author       : betterme
+# @WeChat       : meutils
+# @Software     : PyCharm
+# @Description  : 
+

meutils/apis/audio/deepinfra.py

@@ -0,0 +1,59 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# @Project      : AI.  @by PyCharm
+# @File         : deepinfra
+# @Time         : 2024/11/26 13:57
+# @Author       : betterme
+# @WeChat       : meutils
+# @Software     : PyCharm
+# @Description  : 
+
+from openai import AsyncOpenAI
+
+from meutils.pipe import *
+from meutils.decorators.retry import retrying
+from meutils.io.files_utils import to_bytes
+from meutils.config_utils.lark_utils import get_next_token_for_polling
+from meutils.notice.feishu import send_message as _send_message, AUDIO
+from meutils.schemas.openai_types import STTRequest
+
+FEISHU_URL = "https://xchatllm.feishu.cn/sheets/GYCHsvI4qhnDPNtI4VPcdw2knEd?sheet=6lr4oi"
+
+BASE_URL = os.getenv("DEEPINFRA_BASE_URL")
+
+DEFAULT_MODEL = "openai/whisper-large-v3-turbo"
+MODELS = {
+    "whisper-large-v3": "openai/whisper-large-v3",
+    "whisper-large-v3-turbo": "openai/whisper-large-v3-turbo",
+    "whisper-1": "distil-whisper/distil-large-v3"
+}
+
+send_message = partial(
+    _send_message,
+    title=__name__,
+    url=AUDIO
+)
+
+
+@retrying(max_retries=3, title=__name__)
+async def asr(request: STTRequest, api_key: Optional[str] = None):
+    api_key = api_key or await get_next_token_for_polling(FEISHU_URL)
+
+    request.model = MODELS.get(request.model, DEFAULT_MODEL)
+    request.file = ('_.mp3', await to_bytes(request.file))
+
+    data = request.model_dump(exclude_none=True)
+
+    client = AsyncOpenAI(base_url=BASE_URL, api_key=api_key)
+    response = await client.audio.transcriptions.create(**data)
+    return response
+
+
+if __name__ == '__main__':
+    request = STTRequest(
+        model="whisper-1",
+        file=Path("/Users/betterme/PycharmProjects/AI/MeUtils/meutils/ai_audio/asr/audio.mp3").read_bytes()
+    )
+
+    arun(asr(request))
+    # print(MODELS.values())