eternalblue 1.2.1

A diarization package

EternalBlue is a project for audio diarization and transcription. It separates speakers in a .wav audio file and returns the diarized and transcribed audio.

Description

EternalBlue is a powerful tool for audio analysis, capable of identifying and separating different speakers in a .wav audio file. Additionally, it performs complete transcription of the spoken content, facilitating the analysis and interpretation of conversations.

Installation

Requirements

• Python 3.8 or higher

Installation Steps

You can install EternalBlue via pip:

pip install eternalblue

Usage

from eternalblue import EternalBlue
eternal_blue = EternalBlue('', 'portuguese')
eternalresponse = eternal_blue.diarize(audio, 2)

Here’s an example of how to use the EternalBlue class in a Flask application:

from eternalblue import EternalBlue

@app.route('/diarization', methods=['POST'])
def diarization():
    try:
        eternalresponse = {}
        if 'audio' in request.files:
            audio_file = request.files['audio']
            if audio_file.filename != '':
                save_path = os.path.join(AUDIO_PATH, audio_file.filename)
                audio_file.save(save_path)
                eternal_blue = EternalBlue('', 'portuguese')
                eternalresponse = eternal_blue.diarize(save_path, 2)
        return eternalresponse, 200
    except Exception as e:
        print(f"Error: {e}")
        return str(e), 500

Description

The EternalBlue class requires a Hugging Face API key, the language, and optionally the Whisper model in its constructor. The diarize method expects the path to the .wav audio file and the number of speakers.

Constructor Parameters

• api_key: The Hugging Face API key as a string.

• language: The language of the audio as a string.

• whisper_model (optional): The Whisper model to use.

Method diarize

• audio_file_path: The path to the .wav audio file as a string.

• num_speakers: The number of speakers as an integer.

Expected Return

The diarize method returns a list of dictionaries containing metadata and transcriptions, formatted as follows:

[
    {
        "metadata": [
            {
                "audioname": "tester.wav",
                "time_type": "seconds"
            }
        ]
    },
    {
        "end": 5.984999999999999,
        "speaker_id": "SPEAKER_01",
        "start": 1.299,
        "text": " Is your name Bento Ferreira, then? Are you Brazilian? Yes, sir."
    },
    {
        "end": 5.934,
        "speaker_id": "SPEAKER_00",
        "start": 5.323,
        "text": " Yes, sir."
    },
    {
        "end": 8.498,
        "speaker_id": "SPEAKER_01",
        "start": 6.851,
        "text": " military police officer, That's it ?."
    },
    {
        "end": 13.829,
        "speaker_id": "SPEAKER_00",
        "start": 9.856,
        "text": " Thank God, after 35 years in the reserve."
    },
    {
        "end": 16.613,
        "speaker_id": "SPEAKER_00",
        "start": 15.645,
        "text": " Mr. Miguel. Miguel! "
    },
    {
        "end": 19.618000000000002,
        "speaker_id": "SPEAKER_00",
        "start": 16.969,
        "text": " Any relation to  James ?"
    },

]

This JSON structure contains the following information for each segment: - metadata: General information about the audio file. - speaker_id: The identifier for the speaker. - start: The start time of the speech segment in seconds. - end: The end time of the speech segment in seconds. - text: The transcribed text of the speech segment.