IndicConformer Open-source Automatic Speech Recognition Model - Free Deployment, Accurately Recognize Tamil Speech

Indicconformer Stt Ta Hybrid Ctc Rnnt Large

Developed by ai4bharat

IndicConformer is a Conformer automatic speech recognition model based on hybrid CTC-RNNT architecture, specifically optimized for Tamil

Speech Recognition OtherOpen Source License:MIT #Tamil speech recognition #Hybrid CTC-RNNT architecture #Conformer large model

Downloads 799

Release Time : 9/5/2024

Model Overview

This is an automatic speech recognition (ASR) model supporting Tamil, using Conformer-Large architecture and hybrid CTC-RNNT decoder, capable of converting 16KHz mono audio into text

Model Features

Hybrid decoding architecture

Supports both CTC and RNNT decoding methods, allowing flexible selection based on requirements

Large model capacity

Conformer-Large architecture with 120 million parameters, providing high-accuracy speech recognition

Tamil optimization

Speech recognition model specifically trained and optimized for Tamil

Model Capabilities

Tamil speech recognition

Audio to text conversion

Supports 16KHz mono audio processing

Use Cases

Speech transcription

Automated meeting minutes

Automatically convert Tamil meeting recordings into text transcripts

Improves meeting documentation efficiency and reduces manual transcription time

Media subtitle generation

Automatically generate subtitles for Tamil video content

Enhances video content accessibility and simplifies subtitle production workflow

Voice assistants

Tamil voice interaction

Provide voice control interfaces for Tamil-speaking users

Enables natural language interaction and improves user experience

🚀 IndicConformer

IndicConformer is a Hybrid CTC - RNNT conformer ASR (Automatic Speech Recognition) model, which can effectively transcribe speech into text.

🚀 Quick Start

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

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

✨ Features

Language

This model supports Tamil.

Input

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

Output

For a given audio sample, the model provides transcribed speech as a string.

🔧 Technical Details

Model Architecture

This model is a conformer - Large model. It uses an encoder with 120M parameters and a hybrid CTC - RNNT decoder. The model has 17 conformer blocks, and the model dimension is 512.

📦 Installation

To load, train, fine - tune or play with the model, you need to install AI4Bharat NeMo. Use the following 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_ta_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. Run the following command in your terminal to convert the audio 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='ta')[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='ta')[0]
print(rnnt_text)

📄 License

This project is under the MIT license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご