app.py

import torch
# from PIL import Image
import gradio as gr
import pytube as pt
from transformers import pipeline

MODEL_NAME = "openai/whisper-large-v3"

device = 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)


all_special_ids = pipe.tokenizer.all_special_ids
transcribe_token_id = all_special_ids[-5]
translate_token_id = all_special_ids[-6]


def transcribe(microphone, file_upload, task):
    warn_output = ""
    if (microphone is not None) and (file_upload is not None):
        warn_output = (
            "警告：您已经上传了一个音频文件并使用了麦克录制。"
            "录制文件将被使用上传的音频将被丢弃。\n"
        )

    elif (microphone is None) and (file_upload is None):
        return "错误: 您必须使用麦克风录制或上传音频文件"

    file = microphone if microphone is not None else file_upload

    pipe.model.config.forced_decoder_ids = [[2, transcribe_token_id if task=="transcribe" else translate_token_id]]

    # text = pipe(file, return_timestamps=True)["text"]
    text = pipe(file, return_timestamps=True)
    #trans to SRT
    text= convert_to_srt(text)
    
    return warn_output + text


def _return_yt_html_embed(yt_url):
    video_id = yt_url.split("?v=")[-1]
    HTML_str = (
        f'<center> <iframe width="500" height="320" src="https://www.youtube.com/embed/{video_id}"> </iframe>'
        " </center>"
    )
    return HTML_str


def yt_transcribe(yt_url, task):
    yt = pt.YouTube(yt_url)
    html_embed_str = _return_yt_html_embed(yt_url)
    stream = yt.streams.filter(only_audio=True)[0]
    stream.download(filename="audio.mp3")

    pipe.model.config.forced_decoder_ids = [[2, transcribe_token_id if task=="transcribe" else translate_token_id]]

    text = pipe("audio.mp3",return_timestamps=True)
    # text = pipe("audio.mp3",return_timestamps=True)["text"]
    #trans to SRT
    text= convert_to_srt(text)

    return html_embed_str, text

# Assuming srt format is a sequence of subtitles with index, time range and text
def convert_to_srt(input):
  output = ""
  index = 1
  for chunk in input["chunks"]:
    start, end = chunk["timestamp"]
    text = chunk["text"]
    if end is None:
      end = "None"
    # Convert seconds to hours:minutes:seconds,milliseconds format
    start = format_time(start)
    end = format_time(end)
    output += f"{index}\n{start} --> {end}\n{text}\n\n"
    index += 1
  return output

# Helper function to format time
def format_time(seconds):
  if seconds == "None":
    return seconds
  hours = int(seconds // 3600)
  minutes = int((seconds % 3600) // 60)
  seconds = int(seconds % 60)
  milliseconds = int((seconds % 1) * 1000)
  return f"{hours:02}:{minutes:02}:{seconds:02},{milliseconds:03}"

demo = gr.Blocks()

mf_transcribe = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.inputs.Audio(source="microphone", type="filepath", optional=True),
        gr.inputs.Audio(source="upload", type="filepath", optional=True),
        gr.inputs.Radio(["transcribe", "translate"], label="Task", default="transcribe"),
    ],
    outputs="text",
    layout="horizontal",
    theme="huggingface",
    title="Audio-to-Text-SRT 自动生成字幕",
    description=(
        "直接在网页录音或上传音频文件，加入Youtube连接，轻松转换为文字和字幕格式! 本演示采用"
        f" 模型 [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) 和 🤗 Transformers 转换任意长度的"
        "音视频文件!使用GPU转换效率会大幅提高，大约每小时 $0.6 约相当于人民币 5 元。如果您有较长内容,需要更快的转换速度，请私信作者微信 1259388，并备注“语音转文字”。"
    ),
    allow_flagging="never",
)

yt_transcribe = gr.Interface(
    fn=yt_transcribe,
    inputs=[
        gr.inputs.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
        gr.inputs.Radio(["转译", "翻译"], label="Task", default="transcribe")
    ],
    outputs=["html", "text"],
    layout="horizontal",
    theme="huggingface",
    title="Audio-to-Text-SRT 自动生成字幕",
    description=(
        "直接在网页录音或上传音频文件，加入Youtube连接，轻松转换为文字和字幕格式! 本演示采用"
        f" 模型 [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) 和 🤗 Transformers 转换任意长度的"
        "音视频文件!使用GPU转换效率会大幅提高，大约每小时 $0.6 约相当于人民币 5 元。如果您有较长内容,需要更快的转换速度，请私信作者微信 1259388，并备注“语音转文字”。"
    ),
    allow_flagging="never",
)

#  # Load the images
# image1 = Image("wechatqrcode.jpg")
# image2 = Image("paypalqrcode.png")
    
#     # Define a function that returns the images and captions
# def display_images():
#     return image1, "WeChat Pay", image2, "PayPal"
    
with demo:
    gr.TabbedInterface([mf_transcribe, yt_transcribe], ["转译音频成文字", "YouTube转字幕"])
       
    # Create a gradio interface with no inputs and four outputs
    # gr.Interface(display_images, [], [gr.outputs.Image(), gr.outputs.Textbox(), gr.outputs.Image(), gr.outputs.Textbox()], layout="horizontal").launch()

demo.launch(enable_queue=True)