Indicconformer Stt Ur Hybrid Ctc Rnnt Large

🚀 IndicConformer

IndicConformer is a Hybrid CTC - RNNT conformer ASR (Automatic Speech Recognition) model, designed to transcribe Urdu speech.

🚀 Quick Start

To start using the IndicConformer model, you first need to install AI4Bharat NeMo. We recommend the following installation command:

git clone https://github.com/AI4Bharat/NeMo.git && cd NeMo && git checkout nemo-v2 && bash reinstall.sh

✨ Features

• Language Support: Specifically designed for Urdu language transcription.
• Input Format: Accepts 16000 KHz Mono - channel Audio (wav files).
• Output: Provides transcribed speech as a string for a given audio sample.

📦 Installation

To load, train, fine - tune or play with the model, install AI4Bharat NeMo using the command:

git clone https://github.com/AI4Bharat/NeMo.git && cd NeMo && git checkout nemo-v2 && bash reinstall.sh

💻 Usage Examples

Basic Usage

First, download and load the model from Huggingface:

import torch
import nemo.collections.asr as nemo_asr

model = nemo_asr.models.ASRModel.from_pretrained("ai4bharat/indicconformer_stt_ur_hybrid_rnnt_large")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.freeze() # inference mode
model = model.to(device) # transfer model to device

Get an audio file ready by converting it to 16000 Hz and monochannel:

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

Advanced Usage

Inference using CTC decoder

model.cur_decoder = "ctc"
ctc_text = model.transcribe(['sample_audio_infer_ready.wav'], batch_size=1,logprobs=False, language_id='ur')[0]
print(ctc_text)

Inference using RNNT decoder

model.cur_decoder = "rnnt"
rnnt_text = model.transcribe(['sample_audio_infer_ready.wav'], batch_size=1, language_id='ur')[0]
print(rnnt_text)

📚 Documentation

Model Architecture

This model is a conformer - Large model, consisting of 120M parameters, as the encoder, with a hybrid CTC - RNNT decoder. The model has 17 conformer blocks with 512 as the model dimension.

Input

This model accepts 16000 KHz Mono - channel Audio (wav files) as input.

Output

This model provides transcribed speech as a string for a given audio sample.

📄 License

This project is licensed under the MIT license.