Neura Market
whisper-ctranslate2 logo

whisper-ctranslate2

Free

A Whisper CLI client compatible with the original OpenAI client, using CTranslate2 for faster inference. [#opensource](https://github.com/Softcatala/whisper-ctranslate2)

FreeFree tier
Type
Open Source
Company
Softcatala
Visit Website GitHub (40 stars)

About whisper-ctranslate2

whisper-ctranslate2 is a command line client for OpenAI's Whisper speech recognition model, built on CTranslate2 and Faster-whisper for significantly faster inference. It achieves up to 4x speed improvement over the original OpenAI Whisper at the same accuracy, and up to 16x with batched inference. The tool maintains full CLI compatibility with the original OpenAI client, enabling easy migration. It includes a Docker image with pre-loaded models, supports speaker diarization (speaker identification), voice activity detection (VAD) filtering for improved quality, live transcription from microphone, and color-coded confidence visualization. Users can also load their own fine-tuned Whisper models. The package is installable via pip and works on both CPU and GPU (NVIDIA CUDA).

Key Features

4x faster than OpenAI Whisper at same accuracy (up to 16x with batched inference)
Full command-line compatibility with original OpenAI Whisper client
Docker image with pre-loaded models (small, medium, large-v2)
Speaker Diarization (speaker identification)
Voice Activity Detection (VAD) filter for improved transcription quality
Support for loading custom fine-tuned Whisper models
Live transcription from microphone input
Color-coded confidence visualization
Batched inference for additional speed gains
CPU and GPU support via CTranslate2 (NVIDIA CUDA, Intel MKL, oneDNN, etc.)

Pros & Cons

Pros
  • Up to 4x faster than original OpenAI Whisper while maintaining accuracy
  • Fully compatible with existing OpenAI Whisper CLI scripts and workflows
  • Docker image simplifies deployment with models included
  • Includes speaker diarization for multi-speaker scenarios
  • Voice Activity Detection reduces false positives and improves quality
  • Supports live microphone transcription
  • Offers batched inference for additional speed (up to 16x)
  • Open source with permissive license
Cons
  • Translate task only supports English as the target language
  • Requires NVIDIA GPU for optimal performance (CPU is slower)
  • No built-in support for non-Whisper models
  • Docker image size may be large due to multiple model variants

Best For

Audio transcription with high speed and accuracySpeech-to-text translation (source language to English)Meeting and lecture transcription with speaker identificationReal-time live captioning via microphone inputBatch processing of multiple audio files for media or researchIntegration into pipelines requiring Whisper-compatible CLI

FAQ

How do I install whisper-ctranslate2?
Install via pip: pip install whisper-ctranslate2. Alternatively, use the prebuilt Docker image: docker pull ghcr.io/softcatala/whisper-ctranslate2:latest
How much faster is it compared to OpenAI Whisper?
It is up to 4x faster at the same accuracy when using CTranslate2. With batched inference enabled (--batched True), speed can increase up to 16x.
Does it support GPU?
Yes, it supports GPU execution with NVIDIA CUDA (cuBLAS 11.x required). It also supports CPU with multiple backends (Intel MKL, oneDNN, OpenBLAS, Ruy, Apple Accelerate).
Can I use my own fine-tuned Whisper model?
Yes, you can load your own fine-tuned Whisper models with the tool.