woncomp/malabar
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add dummy line THE END

Browse Source
This commit is contained in:
Joshua Zhang
2026-03-05 17:47:54 +08:00
parent 323d14301d
commit f161e29c7e
11 changed files with 64 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
_speakers/S01E02_speakers.txt
View File

@@ -35,3 +35,4 @@
[03:44](Malabar) Thank you, sun.
[03:47](Narrator) The inside of our star.
[03:48](Narrator) The sun is made of atoms which bump into each other, merge, and produce heat.
[04:00](Narrator) THE END

1
_speakers/S01E03_speakers.txt
View File

@@ -68,3 +68,4 @@
[03:46](Narrator) A long time ago, a very big rock hit the Earth.
[03:50](Narrator) Projecting piles of rocks and dust.
[03:52](Narrator) They were pulled together and formed the moon.
[03:55](Narrator) THE END

1
_speakers/S01E04_speakers.txt
View File

@@ -58,3 +58,4 @@
[03:45](Narrator) Comets are balls made of ice and dust.
[03:48](Narrator) They travel very fast and they're very old.
[03:50](Narrator) When they pass close to the sun, they warm up and leave a tail made of vapor and dust.
[03:55](Narrator) THE END

1
_speakers/S02E02_speakers.txt
View File

@@ -87,3 +87,4 @@
[03:43](Malabar) Good night and sweet dreams, Giraffe.
[03:46](Volcanoes) Long ago, the surface of Mars shook so much that immense canyons were formed.
[03:51](Volcanoes) It's called the Labyrinth of the Night, and it ends in the Valles Marineris.
[03:55](Narrator) THE END

6
_translated/S01E02_translated.json
View File

@@ -220,5 +220,11 @@
"speaker": "Narrator",
"english": "The sun is made of atoms which bump into each other, merge, and produce heat.",
"chinese": "太阳是由原子组成的，它们会相互碰撞、合并，并产生热量。"
},
{
"timestamp": "[04:00]",
"speaker": "Narrator",
"english": "THE END",
"chinese": "THE END"
}
]

6
_translated/S01E03_translated.json
View File

@@ -418,5 +418,11 @@
"speaker": "Narrator",
"english": "They were pulled together and formed the moon.",
"chinese": "它们被引力拉到一起，就形成了月亮。"
},
{
"timestamp": "[03:55]",
"speaker": "Narrator",
"english": "THE END",
"chinese": "THE END"
}
]

6
_translated/S01E04_translated.json
View File

@@ -358,5 +358,11 @@
"speaker": "Narrator",
"english": "When they pass close to the sun, they warm up and leave a tail made of vapor and dust.",
"chinese": "当它们靠近太阳时，就会变热，留下由蒸汽和尘埃组成的尾巴。"
},
{
"timestamp": "[03:55]",
"speaker": "Narrator",
"english": "THE END",
"chinese": "THE END"
}
]

6
_translated/S02E02_translated.json
View File

@@ -532,5 +532,11 @@
"speaker": "Volcanoes",
"english": "It's called the Labyrinth of the Night, and it ends in the Valles Marineris.",
"chinese": "它叫做“黑夜迷宫”，一直延伸到水手谷。"
},
{
"timestamp": "[03:55]",
"speaker": "Narrator",
"english": "THE END",
"chinese": "THE END"
}
]

15
step2_format.py
View File

@@ -293,6 +293,9 @@ def format_lines(transcript_data: Dict[str, Any]) -> str:
if joined_song:
lines.append(f"[00:01](Song) {joined_song}")
# Track the last utterance for calculating THE END timestamp
last_utt = None
# Format remaining lines (skip those within first 15s as they're in the joined song)
for utt in merged:
# Skip utterances within opening song window (they're already included in joined_song)
@@ -313,6 +316,18 @@ def format_lines(transcript_data: Dict[str, Any]) -> str:
timestamp = format_timestamp(utt.get("start", 0))
lines.append(f"{timestamp}(Speaker {speaker}) {text}")
last_utt = utt
# Add dummy "THE END" line after the last line
# Calculate timestamp based on the duration of the last line
if last_utt:
last_start = last_utt.get("start", 0)
last_end = last_utt.get("end", 0)
duration = last_end - last_start
# THE END timestamp = last line start + duration (same as last line's end time)
the_end_time = last_start + duration
the_end_timestamp = format_timestamp(the_end_time)
lines.append(f"{the_end_timestamp}(Narrator) THE END")
return '\n'.join(lines)

27
step3_infer_speakers.py
View File

@@ -98,9 +98,8 @@ def get_llm_config() -> Tuple[str, str]:
def parse_lines(lines_text: str) -> List[Tuple[str, str, str]]:
"""Parse formatted lines. Returns list of (timestamp, speaker_label, text)."""
# Pattern to match both (Speaker X) and (Song) formats
# Speaker "Song" is reserved for the opening song
pattern = r'^(\[\d{2}:\d{2}\])\((Speaker [A-Z?]|Song)\) (.+)$'
# Pattern to match speaker formats: (Speaker X), (Song), (Narrator), etc.
pattern = r'^(\[\d{2}:\d{2}\])\(([^)]+)\) (.+)$'
result = []
for line in lines_text.strip().split('\n'):
@@ -113,9 +112,11 @@ def parse_lines(lines_text: str) -> List[Tuple[str, str, str]]:
timestamp = match.group(1)
speaker_raw = match.group(2)
text = match.group(3)
# Normalize: "Speaker X" -> "X", "Song" -> "Song"
# Normalize: "Speaker X" -> "X", "Song" -> "Song", "Narrator" -> "Narrator"
if speaker_raw == "Song":
speaker = "Song"
elif speaker_raw == "Narrator":
speaker = "Narrator"
else:
# Extract letter from "Speaker X"
speaker = speaker_raw.replace("Speaker ", "")
@@ -205,8 +206,8 @@ def identify_malabar(dialogue_lines: List[Tuple[str, str, str]],
# Output ALL lines in chronological order (preserving original order)
samples = []
for ts, spk, text in dialogue_lines:
# Skip Song speaker for Malabar identification
if spk == "Song":
# Skip Song and Narrator speakers for Malabar identification
if spk in ("Song", "Narrator"):
continue
# Only include speakers we're trying to identify
if spk in speakers:
@@ -269,8 +270,8 @@ def identify_speaker(speaker: str,
elif spk in known_names:
# Known speaker - use real name
full_dialogue_lines.append(f'{known_names[spk]}: "{text}"')
elif spk == "Song":
full_dialogue_lines.append(f'Song: "{text}"')
elif spk in ("Song", "Narrator"):
full_dialogue_lines.append(f'{spk}: "{text}"')
else:
# Unknown speaker - keep as Speaker X
full_dialogue_lines.append(f'Speaker {spk}: "{text}"')
@@ -339,9 +340,9 @@ def process_lines_file(input_path: Path, client: OpenAI, model: str, force: bool
print(" No valid lines found!")
return None
# Get unique speakers (excluding "Song" - already known)
# Get unique speakers (excluding "Song" and "Narrator" - already known)
all_speakers = set(speaker for _, speaker, _ in lines)
speakers_to_identify = [s for s in all_speakers if s != "Song"]
speakers_to_identify = [s for s in all_speakers if s not in ("Song", "Narrator")]
print(f" Speakers to identify: {', '.join(sorted(speakers_to_identify))}")
@@ -431,9 +432,9 @@ def apply_speaker_names(lines: List[Tuple[str, str, str]], mapping: Dict[str, st
result_lines = []
for timestamp, speaker, text in lines:
# "Song" speaker is already correctly labeled - pass through unchanged
if speaker == "Song":
speaker_name = "Song"
# "Song" and "Narrator" speakers are already correctly labeled - pass through unchanged
if speaker in ("Song", "Narrator"):
speaker_name = speaker
else:
speaker_name = mapping.get(speaker, f"Speaker_{speaker}")
result_lines.append(f"{timestamp}({speaker_name}) {text}")

6
step4_translate.py
View File

@@ -247,6 +247,12 @@ def translate_file(input_path: Path, client: OpenAI, model: str) -> Path:
skipped_count += 1
continue
# Skip dummy "THE END" line (added by step 2 for timing purposes)
if english.strip() == "THE END":
line_data["chinese"] = english # Keep as-is
skipped_count += 1
continue
cached = get_cached_translation(english)
if cached:
line_data["chinese"] = cached