Using audio transcription

Invite-only access

Access to Cleura AI services is currently invite-only.

Of the on-demand models we provide, faster-whisper-large-v3 is optimized for speech-to-text (STT) applications, also known as audio transcription.

More specifically, you may use Cleura Cloud’s OpenAI-compatible API to programmatically upload audio files and get back transcribed texts.

Here’s how you can go about it with Python.

Creating an AI API key

If you do not already have one, create a new AI API key. Make sure the key has access to the faster-whisper-large-v3 model.

Make a note of the bearer token of the new API key.

Using the OpenAI Python library

Following is a Python script (stt-demo.py) that uses the OpenAI library:

from openai import OpenAI

client = OpenAI(
    api_key="<your-bearer-token>",
    base_url="https://ai.cleura.cloud/v1"
)

with open("/path/to/the-audio-file.mp3", "rb") as audio_file:
    transcription = client.audio.transcriptions.create(
        model="faster-whisper-large-v3",
        file=audio_file,
    )

print(transcription)

To test STT, replace the-audio-file.mp3 with an audio file from LibriVox. Use, for instance, The Aurora Borealis in 1719 by Sidney Perley (1858-1928). Directly download the file onto your local computer…

$ cd /tmp
$ curl -O https://dn720704.ca.archive.org/0/items/cb13_weather_1712_librivox/cb013_auroraborealis1719_perley_cc_128kb.mp3
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100 3927k  100 3927k    0     0  1639k      0  0:00:02  0:00:02 --:--:-- 1639k

…and replace /path/to/the-audio-file.mp3 in stt-demo.py with /tmp/cb013_auroraborealis1719_perley_cc_128kb.mp3.

Run the script. After a minute or so, you will get the transcription back.