Whisper API (LAURE tool compatible)

Description

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

Prerequisites

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.

Quickstart

Use L.A.U.R.E

after you have modified the .laure file

laure create && laure push && laure run

OR (for testing and development purposes only)

prerequisites

In Dockerfiles add the following lines

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

Build the container

docker build -t testing-stt .

Run the container

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

Delete the container

docker rm -f stt

API Endpoints

Method	Endpoint	Description
GET	/	Welcome message
POST	/transcribe	Transcribe an audio file
GET	/transcribe	Get the transcription of the audio file

/ (GET)

Welcome message

Return

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

Parameter	Type	Description
success	string	The welcome message

/transcribe POST

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

Parameters

Parameter	Type	Description	Additional Info
model	string	The model to use for the transcription	optional, default is base and you can use base/small/medium
file	file	The audio file to transcribe
secret	string	Secret key to use the API	actually not used

Return

Parameter	Type	Description	Additional Info
id	string	The id of the transcription
success	boolean	True if the transcription was successful	only if success
error	string	The error message if the transcription failed	only if not success

/transcribe GET

Get the transcription of the audio file

Parameters

Parameter	Type	Description	Additional Info
id	string	The id of the transcription

Return

Parameter	Type	Description	Additional Info
text	string	The transcription of the audio file
status	string	The status of the transcription	success/running/failed
success	string	Success is present if the transcription succeeded or is running	only if not failed
error	string	The error message if the transcription failed	only if failed

Examples

Transcribe an audio file

Request

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)

Response

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

Get the transcription of the audio file

Request

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)

Response

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

TODO before it's ready to multiple usage

• 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, ...

References

• I use whisper-cpp

License

MIT