mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-16 09:22:22 +00:00
fix(transcription): preserve literal text in WebVTT cues
This commit is contained in:
parent
168a0055a2
commit
67c28cf9f4
2 changed files with 33 additions and 1 deletions
|
|
@ -3,6 +3,7 @@
|
|||
import unicodedata
|
||||
from collections.abc import Sequence
|
||||
from dataclasses import dataclass
|
||||
from html import escape
|
||||
from itertools import accumulate, groupby
|
||||
from typing import Final
|
||||
|
||||
|
|
@ -210,7 +211,7 @@ def _render_vtt(cues: Sequence[SubtitleCue]) -> str:
|
|||
for cue in cues
|
||||
for line in (
|
||||
f"{_format_timestamp(cue.start_ms, '.')} --> {_format_timestamp(cue.end_ms, '.')}",
|
||||
cue.text,
|
||||
escape(cue.text, quote=False),
|
||||
"",
|
||||
)
|
||||
)
|
||||
|
|
|
|||
|
|
@ -1,3 +1,7 @@
|
|||
from typing import Final
|
||||
|
||||
import pytest
|
||||
|
||||
from litellm.litellm_core_utils.audio_utils.subtitle_utils import (
|
||||
SubtitleToken,
|
||||
_merge_tokens_into_words,
|
||||
|
|
@ -105,6 +109,19 @@ class TestRenderSubtitleTokensAsSrt:
|
|||
|
||||
|
||||
class TestRenderSubtitleTokensAsVtt:
|
||||
@pytest.mark.parametrize(
|
||||
("text", "payload"),
|
||||
(
|
||||
("Use <b>bold</b>.", "Use <b>bold</b>."),
|
||||
("Write & literally.", "Write &amp; literally."),
|
||||
("A --> B", "A --> B"),
|
||||
("\"नमस्ते\" & 'hello'", "\"नमस्ते\" & 'hello'"),
|
||||
),
|
||||
)
|
||||
def test_transcript_text_is_escaped_for_cue_payload(self, text: str, payload: str) -> None:
|
||||
tokens: Final = (SubtitleToken(text=text, start_ms=0, end_ms=1000),)
|
||||
assert render_subtitle_tokens_as_vtt(tokens) == f"WEBVTT\n\n00:00:00.000 --> 00:00:01.000\n{payload}\n"
|
||||
|
||||
def test_single_cue_full_document(self):
|
||||
tokens = (
|
||||
SubtitleToken(text="Hello ", start_ms=0, end_ms=500),
|
||||
|
|
@ -130,6 +147,20 @@ class TestRenderSubtitleTokensAsVtt:
|
|||
|
||||
|
||||
class TestSynthesizeSubtitleDocument:
|
||||
@pytest.mark.parametrize(
|
||||
("response_format", "expected"),
|
||||
(
|
||||
("vtt", "WEBVTT\n\n00:00:00.000 --> 00:00:01.000\n<word> &amp;\n"),
|
||||
("srt", "1\n00:00:00,000 --> 00:00:01,000\n<word> &\n"),
|
||||
),
|
||||
)
|
||||
def test_transcript_escaping_is_specific_to_vtt(self, response_format: str, expected: str) -> None:
|
||||
words: Final = (
|
||||
{"word": "<word>", "start": 0.0, "end": 0.5},
|
||||
{"word": "&", "start": 0.5, "end": 1.0},
|
||||
)
|
||||
assert synthesize_subtitle_document(words, response_format) == expected
|
||||
|
||||
WORDS = [
|
||||
{"word": "Four", "start": 0.4, "end": 0.7, "speaker": "spk:0"},
|
||||
{"word": "score", "start": 0.7, "end": 1.1, "speaker": "spk:0"},
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue