MMnode/modules/asr/subtitle_recognition.py
turtle89431 4ea8431cd0
Some checks are pending
Publish to Comfy registry / Publish Custom Node to registry (push) Waiting to run
init
2026-09-26 11:11:23 -07:00

83 lines
3.0 KiB
Python

from __future__ import annotations
from pathlib import Path
from threading import Lock
from . import qwen_asr, whisper_asr
try:
from ...utils.models import require_qwen_asr_model_dirs, require_whisper_large_v3_model_path
from ...utils.subtitles import smart_split_subtitle_segments
except ImportError:
from utils.models import require_qwen_asr_model_dirs, require_whisper_large_v3_model_path
from utils.subtitles import smart_split_subtitle_segments
SUBTITLE_RECOGNITION_METHODS = ["qwen3-asr", "whisper-large-v3"]
_SUBTITLE_RECOGNITION_LOCK = Lock()
class MissingSubtitleRecognitionDependenciesError(RuntimeError):
def __init__(self, dependencies: list[str]) -> None:
self.dependencies = dependencies
packages = " ".join(dependencies)
super().__init__(
f"Missing Python dependencies: {packages}. Install with: pip install {packages}"
)
def validate_subtitle_recognition_method(model_type: str) -> str:
if model_type not in SUBTITLE_RECOGNITION_METHODS:
raise ValueError(
"model_type must be qwen3-asr or whisper-large-v3"
)
return model_type
def subtitle_recognition_options(data: dict) -> tuple[int, bool]:
max_sentence_length = data.get("max_sentence_length", 20)
if (
isinstance(max_sentence_length, bool)
or not isinstance(max_sentence_length, int)
or not 1 <= max_sentence_length <= 500
):
raise ValueError("max_sentence_length must be an integer between 1 and 500")
unload_model = data.get("unload_model", True)
if not isinstance(unload_model, bool):
raise ValueError("unload_model must be a boolean")
return max_sentence_length, unload_model
def recognize_audio_subtitles(
audio_path: Path,
model_type: str,
max_sentence_length: int = 20,
unload_model: bool = True,
) -> list[dict]:
"""Recognize timestamped subtitle segments using a supported local ASR model."""
model_type = validate_subtitle_recognition_method(model_type)
if model_type == "whisper-large-v3":
model_path = require_whisper_large_v3_model_path()
missing_dependencies = whisper_asr.missing_dependencies()
if missing_dependencies:
raise MissingSubtitleRecognitionDependenciesError(missing_dependencies)
with _SUBTITLE_RECOGNITION_LOCK:
segments = whisper_asr.recognize_subtitle_segments(
audio_path,
model_path,
unload_model,
)
return smart_split_subtitle_segments(segments, max_sentence_length)
asr_model_dir, aligner_model_dir = require_qwen_asr_model_dirs()
missing_dependencies = qwen_asr.missing_dependencies()
if missing_dependencies:
raise MissingSubtitleRecognitionDependenciesError(missing_dependencies)
with _SUBTITLE_RECOGNITION_LOCK:
segments = qwen_asr.recognize_subtitle_segments(
audio_path,
asr_model_dir,
aligner_model_dir,
unload_model,
)
return smart_split_subtitle_segments(segments, max_sentence_length)