Run whisper.cpp as Container

Dockerfile

FROM debian:11 AS build

RUN apt-get update \
 && apt-get install -y libsdl2-dev alsa-utils g++ make wget

RUN mkdir /whisper && \
  wget -q https://github.com/ggerganov/whisper.cpp/tarball/master -O - | \
  tar -xz -C /whisper --strip-components 1

WORKDIR /whisper

ARG model
RUN bash ./models/download-ggml-model.sh "${model}"
RUN make main stream

FROM debian:11 AS whisper

RUN apt-get update \
 && apt-get install -y libsdl2-dev alsa-utils \
 && apt-get clean \
 && rm -rf /var/lib/apt/lists/*

WORKDIR /root

ARG model
RUN mkdir /root/models
COPY --from=build "/whisper/models/ggml-${model}.bin" "/root/models/ggml-${model}.bin"
COPY --from=build /whisper/main /usr/local/bin/whisper
COPY --from=build /whisper/stream /usr/local/bin/stream

transcribe.sh

#!/usr/bin/env bash

set -eu

MODEL="base"
LANG="de"

audio_file="${1}"
file_name="$(basename "${audio_file}")"
wav_file="${audio_file%.*}.wav"

media_dir="$(realpath "$(dirname "${audio_file}")")"
script_dir="$(realpath "$(dirname "${0}")")"

docker build -t whisper-${MODEL} --build-arg model=${MODEL} "${script_dir}"

ffmpeg -i "${audio_file}" -ar 16000 "${wav_file}"

docker run --rm -it \
    -v "${media_dir}":/media \
    whisper \
    whisper \
        --model /root/models/ggml-${MODEL}.bin \
        --language ${LANG} \
        -t 2 \
        --output-txt -nt \
        -f "/media/$(basename "${wav_file}")"

Thanks to build caches, there's basically no overhead in running it every time, but you never have to rebuild yourself when changing things. YMMV, though.