'transcribe.cpp' is a GGML-based transcription library that supports all transcription AI models and can be used as a nearly direct replacement for whisper.cpp.



' transcribe.cpp ' is a

GGML- based C/C++ speech recognition library that supports many of the latest transcription models.

Project - transcribe.cpp
https://workshop.cjpais.com/projects/transcribe-cpp



GitHub - handy-computer/transcribe.cpp: ggml speech-to-text inference for 16+ model families · GitHub
https://github.com/handy-computer/transcribe.cpp

According to the official website, the features of transcribe.cpp are as follows:

As of the time of writing, it supports 20 speech recognition (ASR) model families (over 60 models), and more are planned for the future.
- Acceleration using Vulkan, Metal, CUDA, and TinyBLAS
All models have been numerically validated and WER tests have been conducted.
- Support for streaming transcription and batch transcription
- Whisper.cpp can be replaced almost as is.
Officially supported bindings are available in 4 languages.
Python
JavaScript/TypeScript
Rust
Objective-C/Swift

According to the official GitHub repository, the ASR model families supported by transcribe.cpp are as follows:

Parakeet
Canary
Canary-Qwen
Whisper
GigaAM
Moonshine
Moonshine Streaming
Qwen3-ASR
Cohere Transcribe
・SenseVoice
FunASR Nano
Nemotron Speech Streaming
Nemotron 3.5 ASR Streaming
・Multitalker Parakeet Streaming
• Granite Speech 4 / 4.1
Voxtral
Voxtral Realtime
・MedASR
MOSS Transcribe-Diarize
Sortformer

Regarding installation, it can be installed on various platforms if you have CMake and C/C++ build tools. For example, on Windows, you can build `transcribe.cpp` if the following tools are installed.

• winget (※Required to install other tools)
Git
・Visual Studio Build Tools 2022 (C++)
CMake
[code]
# Git
winget install --id Git.Git --accept-source-agreements --accept-package-agreements

# CMake
winget install --id Kitware.CMake --accept-source-agreements --accept-package-agreements

# MSVC C++ build tools (multi-GB download; the C++ workload is NOT default,
# so it must be requested explicitly via --override)
winget install --id Microsoft.VisualStudio.2022.BuildTools `
--accept-source-agreements --accept-package-agreements `
--override '--quiet --wait --norestart --add Microsoft.VisualStudio.Workload.VCTools --add Microsoft.VisualStudio.Component.VC.Tools.x86.x64 --add Microsoft.VisualStudio.Component.Windows11SDK.22621 --includeRecommended'
[code]


The build process can be completed using the following steps.
[code]
git clone https://github.com/handy-computer/transcribe.cpp
cd transcribe.cpp

cmake -B build

cmake --build build --target transcribe-cli --config Release
[code]


As mentioned in the features section, transcribe.cpp aims to replace whisker.cpp, and while it doesn't yet implement all the features of whisker.cpp at the time of writing, it has reached a level where it can replace most use cases. For more detailed build instructions and usage, please see below.

transcribe.cpp/docs/build-windows.md at main · handy-computer/transcribe.cpp · GitHub
https://github.com/handy-computer/transcribe.cpp/blob/main/docs/build-windows.md

in AI,   Software, Posted by log1c_sh