whisper-asr-webservice

Why we included this project

Whisper-asr-webservice takes OpenAI's Whisper speech-to-text models and exposes them through a straightforward HTTP API, so you do not have to build the service layer yourself. It ships as a Docker image that starts a transcription endpoint on CPU or GPU hardware with a couple of commands, which makes it a fast way for a team to get a working endpoint without writing glue code. Behind the API you can pick which engine actually does the recognition: the original OpenAI Whisper, faster-whisper, or WhisperX. The engines differ in speed and accuracy, and WhisperX adds speaker diarization, so the choice is practical rather than cosmetic. Output comes back as plain text, JSON, SRT, VTT, or TSV with word-level timestamps available, which slots into captioning or meeting-note workflows. FFmpeg support also means it accepts common audio and video files rather than only clean WAV input, which helps when the media is real-world and messy.