Subtitle Alchemy is a Python-based tool that automates subtitle generation for audio and video content using the FunASR Speech-to-Text (STT) model, along with a feature that aligns a pre-written transcript with the audio track for accurate subtitling.
- Automatic Subtitle Creation: Generates subtitles from audio/video using FunASR's STT.
- Transcript Matching (文稿匹配): Aligns manually prepared transcripts with audio for precise subtitle timing.
We recommend using Conda to manage the dependencies. Clone the repository and create a new Conda environment using the provided conda-gpu-subalch-py310.yaml file:
git clone https://github.com/ppmzhang2/sub-alchemy.git
cd sub-alchemy
conda env create -f conda-gpu-subalch-py310.yaml
conda activate py310-gpu-subalchFor CPU-only installations, simply use pip from PyPI:
pip install sub-alchemy[cpu]Transcribe an audio or video and save the sketch. The output will be an npz file saved in the specified directory with the same base name as the input file:
subalch transcribe path_to_your_media output_dir --hotword "热词"Generate the subtitle from the sketch. The output will be saved in the specified directory with the same base name as the input file. The --form option allows you to specify the output format (e.g., srt, ass, vtt, etc.), and the --threshold option sets the minimum silence duration (in milliseconds) to split the subtitles:
subalch generate path_to_sketch.npz output_dir --form srt --threshold 500For transcript alignment, simply provide the path to the sketch and the transcript file:
subalch align path_to_sketch.npz path_to_transcript --form srt --threshold 500- Python 3.10
Pull requests are welcome! For major changes, please open an issue to discuss what you'd like to change.