| --- |
| sdk: gradio |
| sdk_version: 5.34.0 |
| --- |
| # Whisper-WebUI |
| A Gradio-based browser interface for [Whisper](https://github.com/openai/whisper) |
|
|
| # Features |
| - Select the Whisper implementation you want to use between: |
| - [openai/whisper](https://github.com/openai/whisper) |
| - [SYSTRAN/faster-whisper](https://github.com/SYSTRAN/faster-whisper) (used by default) |
| - [Vaibhavs10/insanely-fast-whisper](https://github.com/Vaibhavs10/insanely-fast-whisper) |
| - Generate transcriptions from various sources, including **files** & **microphone** |
| - Currently supported output formats: **csv**, **srt** & **txt** |
| - Speech to Text Translation: |
| - From other languages to English (This is Whisper's end-to-end speech-to-text translation feature) |
| - Translate transcription files using Facebook NLLB models |
| - Pre-processing audio input with [Silero VAD](https://github.com/snakers4/silero-vad) |
| - Post-processing with speaker diarization using the [pyannote](https://huggingface.co/pyannote/speaker-diarization-3.1) model: |
| - To download the pyannote model, you need to have a Huggingface token and manually accept their terms in the pages below: |
| 1. https://huggingface.co/pyannote/speaker-diarization-3.1 |
| 2. https://huggingface.co/pyannote/segmentation-3.0 |
|
|
| # Installation and Running |
|
|
| - ## Run Locally |
|
|
| ### Prerequisite |
| To run this WebUI, you need to have `git`, `python` version 3.8 ~ 3.10 & `FFmpeg`.<BR>If you're not using an Nvida GPU, or using a different `CUDA` version than 12.4, edit the file `requirements.txt` to match your environment. |
| |
| Please follow the links below to install the necessary software: |
| - git : [https://git-scm.com/downloads](https://git-scm.com/downloads) |
| - python : [https://www.python.org/downloads/](https://www.python.org/downloads/) |
| - FFmpeg : [https://ffmpeg.org/download.html](https://ffmpeg.org/download.html) |
| - CUDA : [https://developer.nvidia.com/cuda-downloads](https://developer.nvidia.com/cuda-downloads) |
| |
| After installing `FFmpeg`, make sure to **add** the `FFmpeg/bin` folder to your system `PATH` |
| |
| ### Installation using the script files |
| |
| 1. Download the the repository and extract its contents |
| 2. Run `install.bat` or `install.sh` to install dependencies (It will create a `venv` directory and install dependencies there) |
| 3. Start WebUI with `start-webui.bat` or `start-webui.sh` (It will run `python app.py` after activating the venv) |
| |
| - ## Running with Docker |
|
|
| 1. Install and launch [Docker-Desktop](https://www.docker.com/products/docker-desktop/) |
|
|
| 2. Get the repository |
|
|
| 3. If needed, update the `docker-compose.yaml` to match your environment |
|
|
| 4. Docker commands: |
|
|
| Build the image ( Image is about ~7GB) |
| ```sh |
| docker compose build |
| ``` |
| |
| Run the container |
| ```sh |
| docker compose up |
| ``` |
| |
| 5. Connect to the WebUI with your browser at `http://localhost:7860` |
| |
| # VRAM Usages |
| - This project is integrated with [faster-whisper](https://github.com/guillaumekln/faster-whisper) by default for better VRAM usage and transcription speed.<BR>According to faster-whisper, the efficiency of the optimized whisper model is as follows: |
| | Implementation | Precision | Beam size | Time | Max. GPU memory | Max. CPU memory | |
| |-------------------|-----------|-----------|-------|-----------------|-----------------| |
| | openai/whisper | fp16 | 5 | 4m30s | 11325MB | 9439MB | |
| | faster-whisper | fp16 | 5 | 54s | 4755MB | 3244MB | |
| |
| - Whisper's original VRAM usage table for available models: |
| | Size | Parameters | English-only model | Multilingual model | Required VRAM | Relative speed | |
| |:------:|:----------:|:------------------:|:------------------:|:-------------:|:--------------:| |
| | tiny | 39 M | `tiny.en` | `tiny` | ~1 GB | ~32x | |
| | base | 74 M | `base.en` | `base` | ~1 GB | ~16x | |
| | small | 244 M | `small.en` | `small` | ~2 GB | ~6x | |
| | medium | 769 M | `medium.en` | `medium` | ~5 GB | ~2x | |
| | large | 1550 M | N/A | `large` | ~10 GB | 1x | |
| |
| Note: `.en` models are for English only, and you can use the `Translate to English` option from the other models |
