interspeech

NISQA - Non-Intrusive Speech Quality and TTS Naturalness Assessment

The repository provides links to collections of influential and interesting research papers from top AI conferences, with open-source code to promote reproducibility and provide detailed implementation insights beyond the scope of the article. Stay up to date with the latest advances in AI research!

[InterSpeech 24] FreeV: Free Lunch For Vocoders Through Pseudo Inversed Mel Filter

a deep accent recognition network

Supplementary information and code for INTERSPEECH 2018 paper: Singing voice phoneme segmentation by hierarchically inferring syllable and phoneme onset positions

This repository is the implementation of the paper, "Score-balanced Loss for Multi-aspect Pronunciation Assessment" (Interspeech 2023).

PHO-LID: A Unified Model to Incorporate Acoustic-Phonetic and Phonotactic Information for Language Identification

This repo contains the code for "Voice Disorder Analysis: A Transformer-based Approach", accepted at Interspeech 2024

ITALIC: An ITALian Intent Classification Dataset

Interspeech 2019 experiments

Official code for "Activation Steering for Accent Adaptation in Speech Foundation Models" (Interspeech 2026). Parameter-free accent adaptation via mean-shift steering vectors — no weight updates, consistent WER reductions across 8 accents.

The implementation code for the paper "Gate Activation Signal Analysis for Gated Recurrent Neural Networks and Its Correlation with Phoneme Boundaries"

This repo contains the code for "A Contrastive Learning Approach to Mitigate Bias in Speech Models", best student paper award at Interspeech 2024

Interspeech 2025 implementation for mouse USV classification with pretrained audio models.

Code-Switching Sentence Generation by Generative Adversarial Networks and its Application to Data Augmentation. (Interspeech 2019)

30일치 라디오 방송에서 광고 블록을 audio fingerprinting으로 탐지하고 GPT로 메타데이터 추출

Reproduce the Interspeech 2026 TIMBRE paper: the CRIS layer for cross-lingual speech emotion recognition. 49 layers x 26 corpora x 23 languages, one-command reproduction.