Carleslc/AudioToText

What it solves

AudioToText simplifies the process of converting audio files into written text and translating them. It provides a unified interface to use OpenAI's Whisper for transcription and translation to English, and integrates DeepL for translation into other target languages.

How it works

The project provides a Python-based tool (available as a CLI, Jupyter Notebook, or Google Colab) that leverages the Whisper model to transcribe audio from various languages or translate it directly to English. To support translation into languages other than English, it pipes the transcription through the DeepL API. It can output the results in multiple formats, including plain text (TXT), WebVTT (VTT), SubRip (SRT), TSV, and JSON.

Who it’s for

Content creators, researchers, and developers who need to transcribe audio or generate captions for videos and audio files in multiple languages without needing to complex setup.

Highlights

• Multi-model support: Choose between open-source Whisper models (tiny, base, small, medium, large, turbo) or the OpenAI API.
• Broad language support: Transcribes audio from a wide range of languages and translates to English via Whisper or to various other languages via DeepL.
• Caption generation: Generates VTT and SRT files for use in media players like VLC.
• Flexible deployment: Can be run locally via CLI, in Google Colab for cloud GPU access, or within Jupyter environments.
• Apple Silicon optimization: Supports MLX for faster inference on Mac M1/M2/M3/M4 chips.