This is a simple transcriber API that uses whisper to transcribe audio files.

download model you should download the model from here and put it in the models folder.

ex: data/models/ggml-base.bin and data/models/ggml-small.bin and data/models/ggml-medium.bin.

after you have modified the .laure file

laure create && laure push && laure run

ENV LAURE_HOST="0.0.0.0"
ENV LAURE_PORT="5000"

docker build -t testing-stt .

docker run -d -p 5000:5000 --name stt testing-stt

docker rm -f stt

Welcome message

{
  "success": "Welcome to the Whisper API"
}

Transcribe an audio file to text and return the id of the transcription

application/octet-stream is the only accepted content type for the file, use stream=True with requests.post to send the file if you are using requests in python

Get the transcription of the audio file

with open(file_path, "rb") as audio_file:
  files = {"file": audio_file}
  response = requests.post(url=url, data={
      "model": model
  }, files=files, verify=False, stream=True)
result = response.json()
print(result)

{
  "id": "5f7e3b3e-3b3e-4e3b-7e3f-3b3e4e3b7e3f",
  "success": true
}

result = get_text(url, id)
if "error" in result:
  result = {"status": "failed", "text": result["error"]}
elif result["status"] == "success":
  result = {"status": "success", "text": result["text"]}
else:
  result = {"status": "running"}
print(result)

{
  "text": "The transcription of the audio file",
  "status": "success",
  "success": true
}

• Add a way to change the model
• Add a way to change the language
• maybe a ffmpeg to convert the audio file to the right format
• Add a way to get json output
• Add option like temperature, initial_prompt, ...

• I use whisper-cpp

MIT