Subtitles

python subtitles.py lecture.mp3 te-IN    # writes lecture.srt and lecture.vtt

1
00:00:00,000 --> 00:00:04,180
ఈ రోజు మనం భారత రాజ్యాంగం గురించి
మాట్లాడుకుందాం

"""Subtitles: SRT and WebVTT from a transcript's timed segments.

    python subtitles.py lecture.mp3 te-IN        # writes lecture.srt and lecture.vtt

Needs TWENTYSIX_API_KEY and TWENTYSIX_BASE_URL. Long segments are split into captions of at most two lines and
`max_s` seconds, with time shared out by character count (word-level timing is coming soon).
"""
import math
import os
import sys

from twentysix import Client


def captions(segments, max_chars=42, max_lines=2, max_s=7.0):
    """-> [(start, end, text)]. A segment becomes as many captions as its length and duration need; words are shared
    out evenly by character count and so is the segment's time."""
    out = []
    for s in segments:
        words = s["text"].split()
        if "start" not in s or not words:
            continue
        text, dur = " ".join(words), max(0.0, s["end"] - s["start"])
        n = min(len(words), max(1, math.ceil(len(text) / (max_chars * max_lines)), math.ceil(dur / max_s)))
        groups, cur, done = [], [], 0
        for w in words:
            cur.append(w)
            done += len(w) + 1
            if len(groups) < n - 1 and done >= len(text) * (len(groups) + 1) / n:
                groups.append(" ".join(cur))
                cur = []
        if cur:
            groups.append(" ".join(cur))
        total, t = sum(len(g) for g in groups), s["start"]
        for g in groups:
            end = t + dur * len(g) / total
            out.append((round(t, 3), round(end, 3), wrap(g, max_chars)))
            t = end
    return out


def wrap(text, width):
    """Break a caption into two lines at the space nearest its middle."""
    if len(text) <= width or " " not in text:
        return text
    mid = len(text) // 2
    cut = min((i for i, ch in enumerate(text) if ch == " "), key=lambda i: abs(i - mid))
    return text[:cut] + "\n" + text[cut + 1:]


def ts(sec, sep):
    ms = int(round(sec * 1000))
    return f"{ms // 3600000:02d}:{ms // 60000 % 60:02d}:{ms // 1000 % 60:02d}{sep}{ms % 1000:03d}"


def to_srt(caps):
    return "\n".join(f"{i}\n{ts(a, ',')} --> {ts(b, ',')}\n{text}\n" for i, (a, b, text) in enumerate(caps, 1))


def to_vtt(caps):
    return "WEBVTT\n\n" + "\n".join(f"{ts(a, '.')} --> {ts(b, '.')}\n{text}\n" for a, b, text in caps)


if __name__ == "__main__":
    path, language = sys.argv[1], (sys.argv[2] if len(sys.argv) > 2 else "hi-IN")
    with Client() as client:
        result = client.transcribe_file(path, language, mode="native")
    caps = captions(result["segments"])
    stem = os.path.splitext(path)[0]
    open(stem + ".srt", "w", encoding="utf-8").write(to_srt(caps))
    open(stem + ".vtt", "w", encoding="utf-8").write(to_vtt(caps))
    print(f"{len(caps)} captions → {stem}.srt, {stem}.vtt")