Skip to content

This repository contains my work with ASR and NLP. The repository utilizes insanely fast ASR with Whisper and performs several NLP tasks as well.

Notifications You must be signed in to change notification settings

Kabir5296/Multilingual-ASR

Repository files navigation

Multilingual ASR

This repository contains multilingual ASR for Bengali and English.

Created by A F M Mahfuzul Kabir
mahfuzulkabir.com
https://www.linkedin.com/in/mahfuzulkabir \

How to use:

Initialize:

from DoBanglaSummarize import DoBanglaSummarize
from agent import TranscriberAgent

agent = TranscriberAgent()
summarizer_agent = DoBanglaSummarize()

Transcribe, Create Conversation and Get Keywords:

transcription = agent.get_raw_transcription(audio_path)
conversation = agent.create_conversation(audio_path)
keywords = agent.get_keywords(audio_path)
# you can also define your own set of keywords by passing it throug an arguement
keywords = agent.get_keywords(audio_path, keywords=list_of_keywords)

FastAPI Integration

The repository is now integrated with FastAPI for hosting endpoints.

Endpoint 1, 2 & 3: Transcribe, Converse, Summarize
    Input:  'file': String = File path to the audio file.
            'language' : String = 'bn' for Bengali, 'en' for English
    
    Output: Dictionary =    {'content': 
                                {'filename': Filename, 
                                'transcription': Transcription (string) for endpoint 1, List of list for endpoint 2, Summary (string) for endpoint 3
                                }
                            }
Endpoint 4: Get Keyword
    Input:  'file' : String = File path to the audio file.
            'keyword_str' : String = String with keywords separated with comma ','. Example: 'Apple, Mango, Juice'
            'language' : String = 'bn' for Bengali, 'en' for English

    Output: Dictionary =    {'content': 
                                {'filename': Filename, 
                                'keywords': 
                                    {
                                        'keys': Keywords given in the input,
                                        'count': 
                                            {
                                                'key1' : count,
                                                'key2' : count,
                                                .
                                                .
                                            }
                                    }
                                }
                            }

About

This repository contains my work with ASR and NLP. The repository utilizes insanely fast ASR with Whisper and performs several NLP tasks as well.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages