faster-whisper-GUI: a PySide6 desktop front end for faster-whisper and whisperX
faster_whisper GUI with PySide6
At a glance
- What is it?
- CheshireCC/faster-whisper-GUI wraps faster-whisper, whisperX, Silero VAD and Demucs in a PySide6 desktop application that writes srt, txt, smi, vtt and lrc files. It is the right tool if you want those parameters exposed as widgets, and the wrong one if you need a headless pipeline or a current torch build.
- Who is it for?
- Adopt it if you work on a desktop with an NVIDIA GPU and want faster-whisper, whisperX, Silero VAD and Demucs behind one window, with batch processing and word-level timestamps for lyric files. Do not adopt it for headless servers, CPU-only machines, or any pipeline you want to pin and reproduce, since the dependency set is a snapshot of 2024.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 21 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What faster-whisper-GUI adds on top of the command line
faster-whisper is a Python library. Its README example is a few lines of code, and every parameter is a keyword argument you have to look up. faster-whisper-GUI turns that surface into widgets. The repository describes it as a GUI software of faster-whisper that can transcribe audio or video files to srt, txt, smi, vtt and lrc, and states that it provides all parameters of the VAD model and the whisper model. That claim is the product: not a new transcription engine, but a control panel over two engines.
The audience is narrow and identifiable. Someone who has an NVIDIA GPU, a folder of recordings, and no interest in writing a script per batch. The README also points at the WebVTT output and the word-level timestamp mode for karaoke lyrics, played back in foobar2000 with the ESLyric plugin. That is a consumer-facing use case, not a server one. If your job is to transcribe ten thousand files on a schedule, this is the wrong layer.
The engine stack: faster-whisper, whisperX, Silero VAD and Demucs
The application is a shell around several separate projects, and the README links each one. faster-whisper does the transcription through CTranslate2, which is why CTranslate2>=3.21.0 appears in requirements.txt. whisperX is a second path, listed in the README as newly supported, and it brings its own alignment stage. Silero VAD is exposed as a model whose parameters you can edit, so voice activity detection is a setting rather than a fixed preprocessing step. Demucs handles audio and vocal extraction, which the README abbreviates as AVE.
That stacking is the architecture and also the cost. Four upstream projects mean four sets of assumptions about torch, CUDA and model formats. The README does not describe how the pieces are wired internally, and the repository layout (FasterWhisperGUI.py at the top level, a faster_whisper_GUI package, a separate whisperx directory, a config directory) suggests the GUI code and the whisperX integration are not one module. Treat the app as an orchestrator, not a monolith.
The model story is partly manual. The README links to a Hugging Face search for faster-whisper models and says you can also download and convert models inside the software. It also links a converted large-v3 float32 model, hosted on Hugging Face and on Baidu Netdisk. The dependency on a third-party conversion for the largest model is worth noticing: the app can convert, but the README's own recommended path for large-v3 float32 is a pre-converted download.
Installing faster-whisper-GUI from requirements.txt
There is no packaged installer in the repository. The README does not give an install command, does not mention pip, conda, Docker or a prebuilt exe, and the top-level entries contain no setup.py or pyproject.toml. The only dependency manifest is requirements.txt, so the practical install is a virtual environment plus pip against that file. The torch pins are the part that will decide whether this works on your machine.
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtOn Windows the activation line is .venv\Scripts\activate instead. The file pins torch==1.13.1+cu117 and torchaudio==0.13.1+cu117, which are CUDA 11.7 wheels, alongside faster-whisper==0.10.0, pyside6 > 6.5.0, pyside6-fluent-widgets>=1.3.2, CTranslate2>=3.21.0, joblib==1.2.0, pyAV, ffmpeg-python, pyAudio, nltk and webvtt. There is no CPU-only variant listed, so a machine without a CUDA 11.7 compatible driver is not a supported target as far as the manifest shows.
Once the environment is ready, the entry point is the script at the repository root. The README does not state the launch command, but the file name is the natural one:
python FasterWhisperGUI.pyAfter that, the workflow the README documents is: load, download or convert a model, pick an audio or video file, adjust the faster-whisper and VAD parameters, run the transcription, then review the result and edit timestamps in the built-in result view. Output format is your choice among srt, txt, smi, vtt and lrc. For batch work there is a batch process screen, and a file filter screen for narrowing what the batch picks up.
Where the pinned dependency set becomes a real problem
The requirements file is the clearest limitation. torch==1.13.1+cu117 is a 2023-era build. If your driver or your CUDA toolkit has moved on, or if you need a newer torch for another project in the same environment, you cannot satisfy this file without editing it, and editing it may break the whisperX path. The README does not document a supported upgrade path, a compatibility matrix, or which torch versions the app tolerates.
The release history reinforces this. The most recent release listed is 0.8.5 from 2024-12-08, and the last push to the repository was on 2026-08-25. That gap between a 2024 dependency snapshot and 2026 activity on the default branch is not explained by the README, and nothing in the repository entries describes what changed. If you need a reproducible environment, this is a project you pin by commit and test yourself.
The absence of a documented headless mode is the second limitation. Everything the README shows is a window: a result view for editing timestamps, a settings screen, a file list. There is no CLI described, no server mode, no queue. Automation around this app means driving a GUI, which is a fragile thing to build on. If your requirement is a cron job or a service, use faster-whisper directly and skip this layer.
Finally, the README's user agreement is unusual for an open source project. It is a list of prohibited uses written as a condition of using the software, separate from the AGPL-3.0 licence in the repository. The README does not explain how that agreement interacts with the licence grant, and it is not something the documentation resolves.
faster-whisper-GUI against calling faster-whisper from a script
The honest alternative is not another GUI. It is faster-whisper itself, used as a library. The difference is where the parameters live. In the GUI, VAD settings, model parameters, batch options and output format are widgets with defaults you can see and change without reading documentation. In a script, they are keyword arguments you must know exist. For someone transcribing a lecture once, the GUI is faster to a result. For someone transcribing nightly, the script is the only version that can be version-controlled, tested and scheduled.
whisperX is the other comparison, and it is inside this app rather than beside it. The README lists whisperX support and links the upstream project. Choosing between the faster-whisper path and the whisperX path inside the GUI is a real decision about alignment and word-level timing, and the app lets you make it without installing a second tool. That is the strongest argument for the GUI: it is a single place to compare two engines on the same file.
Demucs is a third case. The README supports it but also links UVR and Demucs-Gui as better options for audio and vocal extraction, with the note more and better AVE. That is a rare and useful admission: the maintainer is telling you the built-in extraction is the convenience path, not the best one. If separation quality is the goal, use a dedicated tool and feed the result back in.
Licence, maintenance and what an upgrade costs you
The repository is licensed AGPL-3.0. That is a strong copyleft licence with a network clause. If you modify the app and let users interact with it over a network, the AGPL's source-availability obligation is the thing to read, and it is not the same as running an unmodified copy locally for yourself. The README's separate user agreement adds conditions on top. Nothing here is legal advice; if you plan to redistribute a build or offer it as a service, have someone qualified read both documents together.
The repository is not archived, and the last push was on 2026-08-25. The most recent tagged release is 0.8.5 from 2024-12-08, so the release cadence and the commit activity are not moving together, and the README does not explain the difference. There is no changelog in the top-level entries and no migration notes, which means an upgrade is an exercise in reading diffs. The configuration lives in fasterWhisperGUIConfig.json and huggingface-config.json at the repository root, so settings survive an upgrade in those files, but the README does not document their schema or promise stability across versions. Budget for a manual re-test of your model, VAD and output settings after any pull.
Editorial conclusion
Adopt it if you work on a desktop with an NVIDIA GPU and want faster-whisper, whisperX, Silero VAD and Demucs behind one window, with batch processing and word-level timestamps for lyric files. Do not adopt it for headless servers, CPU-only machines, or any pipeline you want to pin and reproduce, since the dependency set is a snapshot of 2024. Before relying on it, verify that torch 1.13.1+cu117 installs against your driver, confirm the 0.8.5 release is the one you are running, and read the AGPL-3.0 terms if you plan to distribute the app or a modified build.
Frequently asked questions
How do I install faster-whisper-GUI on Windows or Linux?
There is no installer or exe in the repository. The README gives no install command, so the documented path is the requirements.txt file: create a virtual environment, activate it, and pip install -r requirements.txt, then run the FasterWhisperGUI.py script at the repository root.
Does faster-whisper-GUI have a Mac or Linux build?
The repository does not mention a Mac build, and requirements.txt pins torch==1.13.1+cu117 and torchaudio==0.13.1+cu117, which are CUDA wheels. A CUDA 11.7 compatible GPU is what the manifest assumes, so a Mac is not a target the documentation covers.
What output formats can faster-whisper-GUI write?
The README states it transcribes audio or video files to srt, txt, smi, vtt and lrc. Word-level timestamps are supported in the VTT, LRC and SMI formats, which the README describes for karaoke lyric playback.
Can faster-whisper-GUI run whisperX and Demucs?
Yes. The README lists whisperX support and Demucs model support alongside faster-whisper, and the repository contains a separate whisperx directory. The README also links UVR and Demucs-Gui as better options for audio and vocal extraction.
Where do I download models for faster-whisper-GUI?
The README links a Hugging Face search for faster-whisper models and states that you can also download and convert models inside the software. It separately links a converted large-v3 float32 model on Hugging Face and on Baidu Netdisk.
Is there a Docker image or a headless mode for faster-whisper-GUI?
The README does not mention Docker, a CLI or a headless mode. Everything it documents is a PySide6 desktop window, including the result view, the batch process screen and the settings screen.
Community notes