tdnn

Kaldi-based Korean ASR (한국어 음성인식) open-source project

Time delay neural network (TDNN) implementation in Pytorch using unfold method

PyTorch implementation of the Factorized TDNN (TDNN-F) from "Semi-Orthogonal Low-Rank Matrix Factorization for Deep Neural Networks" and Kaldi

基于PaddlePaddle实现的音频分类，支持EcapaTdnn、PANNS、TDNN、Res2Net、ResNetSE等各种模型，还有多种预处理方法

Time Delayed NN implemented in pytorch

PyTorch implementation of a Time Delay Neural Network (TDNN)

Deep Learning using Neural Network Toolbox + Finance Portfolio Selection with MorningStar

tdnn (time delay neural network) tensorflow implementation

This project partially embodies the state-of-the-art practices in speaker verification technology up until 2020, while attaining the state-of-the-art performance on the VoxCeleb1 test sets.

Steps to prepare the data and train a simple HMM-GMM and TDNN model in Kaldi

Speaker verification - Fusion of spectrogram and raw features (ECAPA TDNN + RawNet2)

📈 Model training, Pre-Training and Fine-Tuning workspace, scaling from a single RTX 3090 (24GB).

An experimental automatic speech recognition (ASR) pipeline for Javanese speech. The project prepares and cleans transcribed audio, creates phone or word alignments with a GMM-HMM workflow, and trains PyTorch TDNN and TDNN-F acoustic models.

Rigorous acoustic and prosodic analysis of the Burmese Daily Dialogue Corpus (BDDC) across 13 Synthetic Speakers , Covering F0 , Speaking Rates , MFCCs , and ECAPA-TDNN embeddings.

Speech command recognition with a TDNN (Conv1D) over MFCCs; noise and pitch-shift augmentation to 19k samples and fine-tuning raised test accuracy from 82% to 85%.