ASR Indic Server

Overview

Automatic Speech Recognition (ASR) for Indian languages using IndicConformer models.

Currently verified for kannada, hindi, tamil, telugu, marathi .

Try the web demo at - https://workshop.dwani.ai with Transcription page

Table of Contents

• Getting Started
• For Development (Local)
  • Prerequisites
  • Steps
• Downloading Translation Models
• Kannada
• Other Languages
  • Malayalam
  • Hindi
• Running with FastAPI Server
• Evaluating Results
• Kannada Transcription Examples
  • Sample 1: kannada_sample_1.wav
  • Sample 3 - Song - 4 minutes Building Docker Image
• Run the Docker Image
• Troubleshooting
• References

Getting Started - Development

For Development (Local)

• Prerequisites: Python 3.10 (compatibility verified)
• Steps:
• Create a virtual environment:

  python -m venv venv
  

• Activate the virtual environment:

  source venv/bin/activate
  

  On Windows, use:

  venv\Scripts\activate
  

• Install dependencies:

  pip install -r requirements.txt
  

  sudo apt install ffmpeg
  

Downloading ASR Models

Models can be downloaded from AI4Bharat's HuggingFace repository:

• https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual
• Log in HuggingFace Account
• Request Access to the model

• https://huggingface.co/docs/hub/security-tokens

  • Get a Read token for your account

  export HF_TOKEN=<YOUR-READ-TOKEN-HERE>
  

For Multi-lingual language supported model

hf download ai4bharat/indic-conformer-600m-multilingual 

Sample Code

For all languages

from transformers import AutoModel
import torchaudio
import torch

# Load the model
model = AutoModel.from_pretrained("ai4bharat/indic-conformer-600m-multilingual", trust_remote_code=True)

# Load an audio file
wav, sr = torchaudio.load("kannada_sample_1.wav")
wav = torch.mean(wav, dim=0, keepdim=True)

target_sample_rate = 16000  # Expected sample rate
if sr != target_sample_rate:
    resampler = torchaudio.transforms.Resample(orig_freq=sr, new_freq=target_sample_rate)
    wav = resampler(wav)

# Perform ASR with CTC decoding
transcription_ctc = model(wav, "kn", "ctc")
print("CTC Transcription:", transcription_ctc)

# Perform ASR with RNNT decoding
transcription_rnnt = model(wav, "kn", "rnnt")
print("RNNT Transcription:", transcription_rnnt)

• Run the Code

  python asr-code.py
  

Alternative examples for Development

For Server Development

Running with FastAPI Server

Run the server using FastAPI with the multilingual model

python src/server/asr_api.py --port 10803 --host 0.0.0.0

Evaluating Results for FastApi Server

You can evaluate the ASR transcription results using curl commands.

Kannada Transcription Examples

Sample 1: kannada_sample_1.wav

• Audio File: samples/kannada_sample_1.wav
• Command:

  curl -X 'POST' 'http://localhost:10803/transcribe/language=kannada' -H 'accept: application/json'   -H 'Content-Type: multipa'Content-Type  multipart/form-data' -F 'file=@samples/kannada_sample_1.wav;type=audio/x-wav'
  

• Expected Output: ಕರ್ನಾಟಕದ ರಾಜಧಾನಿ ಯಾವುದು Translation: "What is the capital of Karnataka"

Sample 2 - Song - 4 minutes

• YT Video- Navaduva Nudiye
• Audio File: samples/kannada_sample_3.wav
• Command:

  curl -X 'POST' \
  'http://localhost:10803/transcribe/language=kannada' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@samples/kannada_sample_3.wav;type=audio/x-wav'
  

• Expected Output: kannada_sample_3_out.md

Note: The ASR does not provide sentence breaks or punctuation (e.g., question marks).

Troubleshooting

• Transcription errors: Verify the audio file is in WAV format, mono, and sampled at 16kHz. Adjust using:

  ffmpeg -i sample_audio.wav -ac 1 -ar 16000 sample_audio_infer_ready.wav -y
  

• Model not found: Download the required models using the hf download commands above.
• Port conflicts: Ensure port 10803 is free when running the FastAPI server.

Demo Video

Watch a quick demo of our project in action! Click the image below to view the video on YouTube.

Contributing

We welcome contributions! Please read the CONTRIBUTING.md file for guidelines on how to contribute to this project.

Also you can join the discord group to collaborate

References

• AI4Bharat IndicConformerASR GitHub Repository
• Nemo - AI4Bharat
• IndicConformer Collection on HuggingFace

For Production (Docker)

• Prerequisites: Docker and Docker Compose
• Steps:

• Start the server:

  export HF_TOKEN="YOUR-HF_TOKEN"
  docker compose -f compose.yaml up -d
  

• Steps to build the Docker Image

  docker build -t dwani/asr-indic-server:latest -f Dockerfile .
  

-->