earshot

earshot

earshot is a powerful local speech-to-text library and command-line interface (CLI) constructed with whisper.cpp. Designed for offline use, this tool provides support for CPU and CUDA acceleration, allowing efficient speech transcription without internet connectivity.

Key Features:

• Offline Functionality: Operates entirely offline after downloading the required models.
• Model Support: The default model is large-v3-turbo, and additional models can be easily managed through the command line.
• Output Formats: Supports multiple output formats including TXT, SRT, VTT, and JSON, catering to various use cases.
• Customizable Performance: Utilize CPU-only or CUDA-powered configurations to optimize performance based on hardware capabilities.
• Easy Command-Line Access: Intuitive commands make it simple to transcribe audio files, translate speech, and more.

Quick Start Example:

To initiate the CLI and transcribe an audio file, use the following commands:

cmake -B build -DCMAKE_BUILD_TYPE=Release -DEARSHOT_CUDA=ON
cmake --build build --parallel 2
./build/earshot --list-models
./build/earshot -l en --format txt,srt lesson.mp3

Additional Usage Examples:

• Display help information:
  earshot --help
• Transcribe an interview audio:
  earshot --device cpu -m small.en interview.m4a
• Translate a speech and convert to VTT format:
  earshot --translate -l fr --format vtt talk.wav
• Output podcast transcription to standard output:
  earshot -p --stdout podcast.mp3

Development and Build Options:

Developers can leverage the stable C API or a RAII C++17 wrapper for integration into their applications:

#include <earshot/earshot.hpp>

earshot::Options options;
options.model_path = "/models/ggml-large-v3-turbo.bin";
options.language = "en";
earshot::Transcriber transcriber(options);
const earshot::Result result = transcriber.transcribe("meeting.mp3");
std::cout << result.text() << '\n';

Built on a foundation of ease and efficiency, earshot strives to deliver high-quality speech-to-text transcription for a variety of applications, from personal projects to professional use. Contributions are welcomed, and users are encouraged to explore the robust features this library offers.