Skip to content

Language Server

What is the Language Server?

A unified API platform for accessing AI models and services at one place

Key Features

  • 🎤 Transcription - Convert audio to text with Google and Azure transcription services
  • 🤖 Completion - Generate chat completions from LLMS
  • 🌐 Translation - Translate text across 11 languages with Google, Azure, Sarvam, and NLLB
  • 🔊 Text-to-Speech - Convert text to audio with Azure, Google, Sarvam, and open-source models
  • 🎚️ Audio Engine - Run ordered preprocessing / analysis pipelines on audio in one request (guide)

Available Services

🎤 Transcription

Convert audio files to text using state-of-the-art speech recognition models

Learn More

🤖 Completion

Generate text completions using advanced language models

Learn More

🌐 Translation

Translate text between 11 Indian and global languages

Learn More

🔊 Text-to-Speech

Convert text to natural-sounding audio in 13+ Indian languages

Learn More

🎚️ Audio Engine

Run configurable pipelines (denoise, duration checks, speech-duration via VAD, and more) with signed output URLs

Learn More

API Endpoints Overview

Method Path Purpose
POST /v1/task/batch/transcription Create a new batch transcription task
POST /v1/task/transcription Create a new single transcription task
POST /v1/task/batch/completion Create a new batch completion task
POST /v1/task/completion Create a new single completion task
POST /v1/task/translation Create a single (sync) translation task
POST /v1/task/text-to-speech Create a single (sync) text-to-speech task
GET /v1/status/{task_id} Get status and result of a specific task
GET /v1/tasks/{user_email} List all tasks created by a user
GET /v1/response/{output_dataset_id} Retrieve the output file of a completed batch task
GET /v1/input_datasets/{user_email} List all input datasets uploaded by a user
GET /v1/services List all services provided by the server
POST /v1/audio_engine/pipeline Run an ordered audio component pipeline (multipart; returns task_id and signed URLs)

Getting Started with Projects and API Keys

Note

These are prerequisites for using the Language Server API

Tip

Heads Up: Remember to replace the current Server-URL in your requests with the new one from the Language Server Team.

What is a Project?

A Project represents a logical grouping of tasks and resources. Each project:

  • ✅ Has a unique project_name
  • ✅ Is associated with specific tags and metadata
  • ✅ Owns a dedicated API key for services

Tip

If you're using the platform for personal or individual use, set project_name to "Individual".

What is an API Key?

An API key is a secure token used to authenticate requests made to the platform:

  • 🔐 Project-specific: Each project has its own API key
  • 🔒 Secure: API keys should be kept secret and not shared between projects
  • ✅ Required: Any API request must include a valid key for the associated project

How to Use

  1. Create a Project (currently not accessible through platform, please contact us for project creation)
  2. Obtain the API Key assigned to the project

Include the following in your request:

Header:
X-API-Key: <your_project_api_key>

{
  "project_name": "your_project_name",
  ...
}

Info

The platform will validate that the project_name matches the API key.

Endpoint Details

1. Create Task

Batch Task Creation

  • Endpoint: POST /v1/task/batch/{task_name}
  • Description: Creates a new task, preprocesses the payload, and publishes it to the job queue
  • Task Types: transcription or completion

Parameters: - task_name (path): Name of the task (e.g., transcription, completion) - payload_data (form, required): JSON string with task details

Responses: - 200: Successful Response - 422: Validation Error

Single Task Creation

  • Endpoint: POST /v1/task/{task_name}
  • Description: Creates a new task, preprocesses the payload, and publishes it to the job queue
  • Task Types: transcription, completion, translation, or text-to-speech

Parameters: - task_name (path): Name of the task (e.g., transcription, completion, translation, text-to-speech) - payload_data (form, required): JSON string with task details - files (form, conditional): Array of files — required for transcription (audio files). Translation and text-to-speech do not require file uploads.

Warning

The total size of all files in a single task upload must not exceed 25 MB.

Responses: - 200: Successful Response - 422: Validation Error

2. Get Task Status and Response

  • Endpoint: GET /v1/status/{task_id}
  • Description: Checks the status of a task. If the task is completed, returns the result; otherwise, returns the current status.

Parameters: - task_id (path, required): ID of the task to check

Responses: - 200: Successful Response - 422: Validation Error

3. List Tasks by User

  • Endpoint: GET /v1/tasks/{user_email}
  • Description: Lists all the tasks created by a user.

Parameters: - user_email (path, required): Email of the user

Responses: - 200: Successful Response - 422: Validation Error

4. List Services

  • Endpoint: GET /v1/services
  • Description: Lists all the services provided by the language server.

Parameters: None

Responses: - 200: Successful Response

Example Usage

Create Task

Basic Request Format:

POST /v1/task/batch/{task_name}
Accept: application/json
X-API-Key: YOUR_API_KEY
Content-Type: multipart/form-data

payload_data={"user_email":"user@example.com","task_type":"BATCH","project_name":"Individual",...additional_fields}
files=@your_file_here

Transcription Example

POST /v1/task/batch/transcription
Accept: application/json
X-API-Key: YOUR_API_KEY
Content-Type: multipart/form-data

payload_data={"user_email":"user@example.com", "project_name":"Individual","task_type":"BATCH","provider":"AZURE","locale":"en-US"}
files=@audio_file.wav

Completion Example

POST /v1/task/batch/completion
Accept: application/json
X-API-Key: YOUR_API_KEY
Content-Type: multipart/form-data

payload_data={"user_email":"user@example.com","project_name":"Individual","task_type":"BATCH","models":["gemini-2.5-flash"]}
files=@requests.jsonl

Translation Example

POST /v1/task/translation
Accept: application/json
X-API-Key: YOUR_API_KEY
Content-Type: multipart/form-data

payload_data={"user_email":"user@example.com","project_name":"Individual","source_language":"en","target_language":"hi","provider":"GOOGLE","text":["Hello, how are you?"]}

Text-to-Speech Example

POST /v1/task/text-to-speech
Accept: application/json
X-API-Key: YOUR_API_KEY
Content-Type: multipart/form-data

payload_data={"user_email":"user@example.com","project_name":"Individual","provider":"AZURE","models":["NEURAL"],"datapoints":{"dp1":{"text":"Hello, this is a test.","speed":"1.0x","lang":"en-IN"}}}

Get Task Status

GET /v1/status/{task_id}

List Tasks by User

GET /v1/tasks/{user_email}

List Services

GET /v1/services

Task Status and Polling

Status Values

All tasks go through the following states:

Status Description
PENDING Task is queued and waiting to be processed
COMPLETED Task finished successfully
FAILED Task encountered an unrecoverable error
PARTIAL_COMPLETE Task completed but some items within it failed

Polling Pattern (Batch Tasks)

Batch tasks (transcription and completion) are processed asynchronously via a queue. After submitting a batch task, poll the status endpoint until the status is terminal:

import time, requests

task_id = "your-task-id"
headers = {"X-API-Key": "YOUR_API_KEY"}

while True:
    r = requests.get(f"https://languageserver.karya.in/v1/status/{task_id}", headers=headers)
    data = r.json()
    if data["status"] in ("COMPLETED", "FAILED", "PARTIAL_COMPLETE"):
        print(data)
        break
    time.sleep(10)  # wait 10 seconds before polling again

Info

Single tasks (translation, text-to-speech, and single transcription/completion) return results directly in the POST response — no polling needed.

Available Models

Transcription Models

Provider Available Models Default Model
Azure base base
Google long, short, telephony, telephony_short long
Sarvam saarika:v2.5 saarika:v2.5
AWS base base

Completion Models

Model Provider Status
gemini-3-flash-preview Vertex AI — Google Default
gemini-2.5-flash Vertex AI — Google Available
gemini-2.5-pro Vertex AI — Google Available
gemini-3.1-pro-preview Vertex AI — Google Available
gemini-3.1-flash-lite-preview Vertex AI — Google Available
gpt-4o Azure OpenAI Available
gpt-4o-mini Azure OpenAI Available
gpt-5 Azure OpenAI Available (single only)
gpt-5-chat Azure OpenAI Available (single only)
gpt-5-mini Azure OpenAI Available (single only)
gpt-5-nano Azure OpenAI Available (single only)
claude-sonnet-4 Vertex AI — Anthropic Available
claude-haiku-4-5 Vertex AI — Anthropic Available
claude-sonnet-4-6 Vertex AI — Anthropic Available
claude-opus-4-6 Vertex AI — Anthropic Available

How to Use Single Task Endpoint

Info

Single task endpoint format: /task/{task_name} where task_name can be 'completion', 'transcription', 'translation', or 'text-to-speech'

Transcription Workflow

  1. Prepare your audio files and a JSON payload with required fields
  2. Make a POST request to the transcription endpoint with the payload and files
  3. Receive a response and status of the task

Completion Workflow

  1. Prepare a .jsonl file with each line as a request
  2. Make a POST request to the completion endpoint with the payload and the .jsonl file
  3. Receive a response and task status

Translation Workflow

  1. Prepare your text as a list of strings and a JSON payload with source_language, target_language, and provider
  2. Make a POST request to the translation endpoint with payload_data as a form field
  3. Receive the translated text directly in the response

Text-to-Speech Workflow

  1. Prepare your datapoints dict (each entry has text, speed, and lang) and a JSON payload with provider and models
  2. Make a POST request to the text-to-speech endpoint with payload_data as a form field
  3. Receive audio download URLs (saas_url) in the response — valid for 24 hours