StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models

Self-Supervised Speech Pre-training and Representation Learning Toolkit

Research and Production Oriented Speaker Verification, Recognition and Diarization Toolkit

A low-bitrate single-codebook 16 / 24 kHz speech codec based on focal modulation

A collections of audio codecs with a standardized API

This repo contains the source code of the first deep learning-base singing voice beat tracking system. It leverages WavLM and DistilHuBERT pre-trained speech models to create vocal embeddings and trains linear multi-head self-attention layers on top of them to extract vocal beat activations. Then, it uses HMM decoder to infer signing beats and t…

A neural speech codec based on discrete WavLM representations

The code for the MAPSS measures for source separation evaluation.

This repository combines `WavLM`, a powerful speech representation model from Microsoft, with `MSDD` (Multi-Scale Diarization Decoder), a state-of-the-art approach for speaker diarization from Nvidia.

In this repository, the wavLM model is used for quality and poor quality data for speaker verification task, and the PyCM library is used for evaluation.

This repository contain the code of the main part of my master thesis degree at Politecnico di Torino in Data science & Engineering

Layer-aware TDNN: Speaker Recognition Using Multi-Layer Features from Pre-Trained Models, to appear in ICAIIC 2026

SOTA method for self-supervised speaker verification leveraging a large-scale pretrained ASR model.

Rethinking Leveraging Pre-Trained Multi-Layer Representations for Speaker Verification, ISCA Interspeech 2025

This repo contains code used in the paper "Characterizing the temporal dynamics of universal speech representations for generalizable deepfake detection"

WavLM Large + RawNetX Speaker Verification Base: End-to-End Speaker Verification Architecture

Acoustic Transformer Models for Audio Classification

This project investigates the performance of different Machine Learning pipelines applied to the task of speech emotion recognition (SER) on the Italian Emozionalmente dataset. Pipelines: MFCC, Wav2Vec and WavLM (both as feature extractors and after fine-tuning), Audio Spectrogram Transformer, cross-linguistic evaluation with f-t WavLM on CREMA-D.

CryCeleb2023 experiments