app.py

import io
import sys
import gradio as gr
import srt
import jiwer

from dataclasses import dataclass
from dataclasses_json import dataclass_json
from datetime import timedelta


@dataclass_json
@dataclass
class ZHTW_Sub:
    start: timedelta
    end: timedelta
    zh: str
    tw: str

def read_srt(p):
    with open(p) as f:
        subs = list(srt.parse(f.read()))
    return subs
   
def merge_sub(subs):
    i = 1
    while i < len(subs):
        ps = subs[i-1]
        s = subs[i]
        if ps.end != s.start:
            i += 1
            continue

        ps.end = s.end
        ps.zh += f" {s.zh}"
        ps.tw += f" {s.tw}"
        subs.pop(i)
    return subs

def merge_sub2(subs, delta):
    i = 1
    while i < len(subs):
        ps = subs[i-1]
        s = subs[i]
        if s.start - ps.end > delta:
            i += 1
            continue

        ps.end = s.end
        ps.zh += f" {s.zh}"
        ps.tw += f" {s.tw}"
        subs.pop(i)
    return subs

def filter_sub(subs):
    buffer = io.StringIO()
    stdout_bak = sys.stdout
    sys.stdout = buffer  # Redirect print to buffer
    
    new_subs = []
    carry_next = False
    for s in subs:
        content = s.content
        if '#' in s.content:
            print('註：標記', s.start, s.end, s.content)
            continue
        
        if '\n' in content:
            print('修：分行', '\\n', s.start, content)
            carry_next = True
            continue #?
        else:
            content = [content]
            
        if len(content) != 1:
            print('註：多行', '\\n', s.start, content)
            print(s.start, s.end)

        tw_all, zh_all = [], []
        for cnt in content:
            if '|' in cnt:
                if len(cnt.split('|')) %2 != 0:
                    print('修：多槓', cnt.split('|'))
                    continue
                tw, zh = cnt.split('|')
                tw, zh = (t.strip() for t in [tw, zh])
    
            else:
                sp = cnt.split()
                if len(sp) %2!=0:
                    print('修：不均', s.start, s.end, sp)
                    continue
                else:
                    mid = len(sp)//2
                    tw, zh = sp[:mid], sp[mid:]
                    tw, zh = (' '.join(t) for t in [tw, zh])
                    if jiwer.cer(tw, zh) > 1:
                        print('註：差距', s.start, s.end, 'tw:', tw, 'zh:', zh)
            tw_all.append(tw)
            zh_all.append(zh)
        if carry_next:
            new_subs[-1].zh += f" {zh}"
            new_subs[-1].tw += f" {tw}"
            new_subs[-1].end = s.end
            carry_next = False
        else:
            new_sub = ZHTW_Sub(s.start, s.end, zh, tw)
            new_subs.append(new_sub)
    sys.stdout = stdout_bak
    return new_subs, buffer

def update_yield():
    buffer = []
    def update_print(inp):
        buffer.append(str(inp))
        return '\n'.join(buffer)
    return update_print

def parse_srt(file):
    if file is None:
        return "No file uploaded."

    upd = update_yield()
    yield upd(file.name)
    subs = read_srt(file.name)
    yield upd(len(subs))
    new_subs, logs = filter_sub(subs)
    yield upd(logs.getvalue())
    yield upd(len(new_subs))
    new_subs = merge_sub(new_subs)
    yield upd(len(new_subs))

    # ep_name = file.name.replace('-dedup', '')
    # ep_name = ep_name.replace('.fix', '')
    total_dur = 0
    for i, it in enumerate(new_subs):
        if (it.end-it.start).total_seconds() > 30:
            yield upd(i)
            yield upd(it.end.total_seconds(), (it.end-it.start).total_seconds(), it.tw)
        total_dur += (it.end-it.start).total_seconds()
    yield upd("可用時長 "+str(timedelta(seconds=int(total_dur))))

with gr.Blocks() as demo:
    gr.Markdown("## SRT File Validator")

    with gr.Column():
        file_input = gr.File(label="Upload .srt File", file_types=[".srt"])
        output_log = gr.Textbox(label="Parsing Log", lines=10, max_lines=120)

    file_input.change(fn=parse_srt, inputs=file_input, outputs=output_log)

demo.launch()