Use WhisperCPP to transcribe audio and video files.
Follow the instructions in the WhisperCPP README to build the binaries for your platform of choice.
Create a .env
file in the root of this project and specify the following paths:
ftmtr_data_root=
ftmtr_whisper_executable=
ftmtr_whisper_model_root=
ftmtr_data_root
corresponds to the directory where audio-only files and JSON files containing the transcription text will be stored temporarily.
ftmtr_whisper_executable
is the full path to the whisper-cli
executable.
ftmtr_whisper_model_root
is the full path to the directory containing WhisperCPP models.
Optionally, the .env
file can also contain the following values:
ftmtr_whisper_timeout=
ftmtr_whisper_model=
ftmtr_whisper_language=
In order to run WhisperCPP on your device, install the library:
poetry install
To run commands inside of the .venv
created:
$(poetry env activate)
To start an openaleph-procrastinate worker that can execute transcription tasks, start the worker from the command line:
PROCRASTINATE_APP=ftm_transcribe.tasks.app procrastinate worker -q transcribe
To defer tasks from other places, use transcribe
as queue name and ftm_transcribe.tasks.transcribe
as the task identifier. These values can also be imported from openaleph-procrastinate.settings.
ftm-transcribe
, (C) 2025 Data and Research Center – DARC
ftm-transcribe
is licensed under the AGPLv3 or later license.