Skip to content

audio_decode.md

Latest commit

 

History

History
56 lines (42 loc) · 1.52 KB

audio_decode.md

File metadata and controls

56 lines (42 loc) · 1.52 KB
Raw

audio_decode

audio_decode is a pre-processing module for ASR requests:

POST /v1/audio/transcriptions
client_encoded_audio_request -> core_wav_audio_request

It accepts MP3 and FLAC input through JSON paths or multipart uploads, decodes the file with miniaudio, writes a temporary WAV, rewrites the request to a core-compatible JSON request, then lets the core ASR handler run normally.

Build

cmake -S . -B build/debug \
  -DCMAKE_BUILD_TYPE=Debug \
  -DAUDIOCPP_BUILD_SERVER_FRONTENDS=ON \
  -DAUDIOCPP_SERVER_FRONTENDS_DIR=/path/to/audio.cpp-server-frontends \
  -DAUDIOCPP_SERVER_FRONTEND_MODULES=audio_decode

audio_decode uses the vendored miniaudio header and does not add a system library dependency.

Open WebUI

Use the standard (full) Open WebUI image, such as ghcr.io/open-webui/open-webui:main, for microphone transcription. Its slim images omit FFmpeg and bypass audio preprocessing, so browser recordings can reach audio.cpp as WebM and receive HTTP 400. This frontend accepts MP3/FLAC in addition to the core's WAV input; it does not decode WebM/Opus.

The full Open WebUI image can transcode these recordings to MP3 before sending them. Leave BYPASS_PYDUB_PREPROCESSING unset or false in Open WebUI. FFmpeg in the audio.cpp container does not perform this client-side conversion.

Request

Multipart upload:

curl http://127.0.0.1:8080/v1/audio/transcriptions \
  -F model=qwen3-asr \
  -F file=@speech.mp3

JSON path:

{
  "model": "qwen3-asr",
  "file": "speech.flac"
}