[ICLR 2025] SOTA discrete acoustic codec models with 40/75 tokens per second for audio language modeling
FunCodec is a research-oriented toolkit for audio quantization and downstream applications, such as text-to-speech synthesis, music generation et.al.
[ICCV 2025] SimVQ: Addressing Representation Collapse in Vector Quantized Models with One Linear Layer
A Survey of Spoken Dialogue Models (60 pages)
unofficial implementation of the High Fidelity Neural Audio Compression
Codebase for the paper 'EncodecMAE: Leveraging neural codecs for universal audio representation learning'
A collections of audio codecs with a standardized API
A TTS app where you can clone the voices of any person you wish.
Token U-Net research for restoring degraded full-mix music in the neural-codec domain.
Academic research project exploring AI music generation - Harvard CS 109B experimental implementation (NOT production-ready)
Experiments sonifying frame-level encodecmae features and encodecmae summary vectors using generative audio models.
Offline removal of neural-codec (RVQ) quantization artifacts from AI-generated music - comb suppression, frozen-noise unfreezing, optional Demucs neural stage
Phone-level speech inpainting: swap or repair individual phones in speech audio
Speech codec prototype for waveform-to-token experimentation with EnCodec, reconstruction analysis, and Apple Silicon validation.
Autoregressive speech prediction with EnCodec and FACodec token language models: 2s context -> 1s future speech, with STOI/PESQ/DNSMOS analysis of where those metrics disagree.
Audio-token compression benchmark for speech ML infra: EnCodec + ASR/Speaker evals, Modal GPU runs, salience/trained selectors, KV-cache serving analysis, and public benchmark artifacts.
Four-stage audio pipeline: split a track into content and style, calibrate a 160-D genre space, and remaster it into a target genre.
To associate your repository with the encodec topic, visit your repo's landing page and select "manage topics."