bhashini-client

bhashini-client is a Python SDK for working with Bhashini inference APIs. The package wraps common pipeline calls behind simple service classes and a unified BhashiniClient, with direct service IDs embedded in each service as requested.

Overview

The SDK currently supports:

• ASR (Automatic Speech Recognition)
• NMT (Neural Machine Translation)
• TTS (Text-to-Speech)
• Transliteration
• Text Language Detection
• OCR
• NER

Each service includes:

• input preprocessing and validation
• hardcoded serviceId inside the service implementation
• safe postprocessing for clean outputs
• graceful fallback to "Invalid input" and "API Error"

Installation

pip install .

Usage

from bhashini_client import BhashiniClient

client = BhashiniClient(api_key="your-api-key")

print(client.asr("https://example.com/audio.wav", "hi"))
print(client.nmt("Hello world", "en", "hi"))
print(client.tts("नमस्ते दुनिया", "hi"))
print(client.transliterate("namaste", "en", "hi"))
print(client.detect_language("नमस्ते दुनिया"))
print(client.ocr("https://example.com/page.png", "hi"))
print(client.ner("OpenAI is in Delhi", "en"))

Services

ASR

Transcribes speech from a remote audio URL.

NMT

Translates text from one language to another.

TTS

Generates speech output and returns a usable audio location or content.

Transliteration

Returns transliterated text with suggestions.

Language Detection

Returns the most likely language code and score.

OCR

Extracts text from printed, scene, or handwritten image URLs.

NER

Extracts named entities as a list of {text, label} dictionaries.

Testing Approach

Tests are written with pytest and cover:

• normal cases
• empty input
• whitespace input
• numeric input
• mixed input
• invalid input type
• API failure fallback

Every test prints Test ID, Input, and Output. A test_results.xlsx workbook is generated automatically at the project root after test execution to help track expected versus actual behavior.

Run tests with:

pytest