mrluyao/whisper-ctranslate2

Whisper command line client compatible with original OpenAI client based on CTranslate2.

Introduction

Whisper command line client compatible with original OpenAI client based on CTranslate2.

It uses CTranslate2 and Faster-whisper Whisper implementation that is up to 4 times faster than openai/whisper for the same accuracy while using less memory.

Goals of the project:

• Provide an easy way to use the CTranslate2 Whisper implementation
• Ease the migration for people using OpenAI Whisper CLI

Installation

Just type:

pip install -U whisper-ctranslate2

Alternatively, the following command will pull and install the latest commit from this repository, along with its Python dependencies:

pip install git+https://github.com/jordimas/whisper-ctranslate2.git

Usage

Same command line that OpenAI whisper.

To transcribe:

whisper-ctranslate2 inaguracio2011.mp3 --model medium

To translate:

whisper-ctranslate2 inaguracio2011.mp3 --model medium --task translate

Additionally using:

whisper-ctranslate2 --help

All the supported options with their help are shown.

CTranslate2 specific options

On top of the OpenAI Whisper command line options, there are some specific options provided by CTranslate2 .

--compute_type {default,auto,int8,int8_float16,int16,float16,float32}

Type of quantization to use. On CPU int8 will give the best performance.

--model_directory MODEL_DIRECTORY

Directory where to find a CTranslate Whisper model, for example a fine-tunned Whisper model. The model should be in CTranslate2 format.

--device_index [DEVICE_INDEX ...]

Device IDs where to place this model on

--vad_filter VAD_FILTER

Enable the voice activity detection (VAD) to filter out parts of the audio without speech. This step is using the Silero VAD model https://github.com/snakers4/silero-vad.

--vad_min_silence_duration_ms VAD_MIN_SILENCE_DURATION_MS

When vad_filter is enabled, audio segments without speech for at least this number of milliseconds will be ignored.

Whisper-ctranslate2 specific options

On top of the OpenAI Whisper and CTranslate2, whisper-ctranslate2 provides some additional specific options:

--print-colors PRINT_COLORS

Adding the --print_colors True argument will print the transcribed text using an experimental color coding strategy based on whisper.cpp to highlight words with high or low confidence:

Contact

Jordi Mas [email protected]