MT (Text-to-Text)¶

The endpoint for the Machine Translation service is https://asr.iitm.ac.in/internal/mt/translate. This API supports the translation services via a LoRA finetuned NLLB model trained at our SPRING lab on Technical domain-specific data from NPTEL.

Request keys¶

Key	Description
`src_language`	source language of the text to be translated in lowercase (eg: `hindi`, `telugu`). Default option: `english`.
`tgt_language`	target language to which the input data is to be translated into. all in lowercase (eg: `sanskrit`, `urdu`). Default option: `english`.
`transcript` (optional)	source text being passed for translation.
`source_vtt` (optional)	data being passed for translation as a string, but with captions as in a `.vtt` file. Each segment/caption of the vtt is translated independently.
`translator_choice`	the translator option being chosen to translate the input. Choose from `iiit-h`, `Finetuned_NLLB`. Default option: `Finetuned_NLLB`.

Response keys¶

Upon successful service of the request, the API returns a JSON response with the following keys:

Key	Description
`status`	success
`mt_out`	If the `transcript` key is passed with some text in the request, `mt_out` would carry the translation of the same.
`translated_vtt`	If the `source_vtt` key is passed with vtt content in the request, `translated_vtt` would carry the translation of each of the captions in the `source_vtt` and returns the translated vtt content.
`transcript`	If the `transcript` key is passed with some text in the request, additional punctuations would have been incorporated into the text and the richer text is returned.
`source_vtt`	If the `source_vtt` key is passed with vtt content in the request, additional punctuations would have been added to each of the captions in the `source_vtt` and the richer source vtt content is returned.

In case of a service failure, the API returns a JSON response with the following keys:

Key	Description
`status`	failure
`reason`	a reason for the failure in serving the request

Language pairs supported by `iiit-h`¶

Language pairs supported by `Finetuned_NLLB`¶

The NLLB model from Meta AI supports 200 languages. Every possible set of language pairs from the below set is supported by the Finetuned_NLLB translator choice.

{'Bengali', 'Gujarati', 'Hindi', 'Kannada', 'Malayalam', 'Marathi', 'Tamil',
'Telugu' }

Usage¶

PythoncURL

import requests

text = "सुप्रभात. आईआईटी मद्रास में आपका स्वागत है। "

payload = {
'src_language': 'hindi',
'tgt_language': 'english',
'transcript': text,
'source_vtt': None,
'translator_choice': 'Finetuned_NLLB'
}

response = requests.post('https://asr.iitm.ac.in/internal/mt/translate', data = payload).json()
print(response)

curl -v -X POST -F 'src_language=english' -F 'tgt_language=telugu' \
-F 'transcript=Welcome to our lab' -F 'translator_choice=iiit-h' \
https://asr.iitm.ac.in/demo/ttt

Web Demo interface for the NLLB model is available at https://asr.iitm.ac.in/demo/ttt

Last update: August 2024
Created: March 24, 2023