audio-transcription

README

Audio Transcription

Speech-to-text transcription using the Zerfoo OpenAI-compatible API server.

How it works

1. Loads a Whisper GGUF model via inference.LoadFile
2. Implements the serve.Transcriber interface to bridge the model to the API server
3. Starts an in-process HTTP server using serve.NewServer with serve.WithTranscriber
4. Sends the audio file to /v1/audio/transcriptions using multipart form upload
5. Prints the transcription result

This demonstrates embedding a full OpenAI-compatible transcription service inside a Go application. The same endpoint works with any OpenAI client library.

Prerequisites

Requires a Whisper-architecture GGUF model file.

Usage

go build -o audio-transcription ./examples/audio-transcription/

# Transcribe an audio file
./audio-transcription --model path/to/whisper.gguf --audio recording.wav

# With language hint
./audio-transcription --model path/to/whisper.gguf --audio recording.mp3 --language en

# With GPU
./audio-transcription --model path/to/whisper.gguf --device cuda --audio recording.wav

Flags

Flag	Default	Description
--model	(required)	Path to a Whisper GGUF model file
--audio	(required)	Path to an audio file (WAV, MP3, FLAC, OGG)
--device	cpu	Compute device: "cpu", "cuda"
--language	(auto)	Optional language hint (e.g., "en", "fr")

Documentation

Overview

Command audio-transcription demonstrates speech-to-text using the Zerfoo OpenAI-compatible API server.

It starts an in-process API server with a Whisper model, then sends an audio file to the /v1/audio/transcriptions endpoint -- the same API that OpenAI clients use. This shows how to embed a full transcription service inside a Go application.

Usage:

go build -o audio-transcription ./examples/audio-transcription/
./audio-transcription --model path/to/whisper.gguf --audio recording.wav