gemelo-ai/vocos

Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis

What it solves

Vocos is a neural vocoder designed to synthesize high-quality audio waveforms from acoustic features (such as mel-spectrograms or EnCodec tokens). It aims to bridge the gap between time-domain and Fourier-based neural vocoders, providing a fast, high-fidelity reconstruction of audio.

How it works

Unlike traditional GAN-based vocoders that model audio samples in the time domain, Vocos generates spectral coefficients. These coefficients are then converted into audio waveforms using an inverse Fourier transform, allowing the model to generate audio in a single forward pass for rapid reconstruction.

Who it’s for

This tool is for researchers and developers building audio synthesis systems, text-to-speech (TTS) pipelines, or integrating with other audio models like Bark.

Highlights

  • Fast Synthesis: Generates waveforms in a single forward pass.
  • Spectral Domain Modeling: Uses spectral coefficients and inverse Fourier transform instead of time-domain modeling.
  • GAN-based Training: Trained using a Generative Adversarial Network objective.
  • Flexible Input: Supports reconstruction from both mel-spectrograms and EnCodec tokens.

Related

  • Project
    janvarev/Irene-Voice-Assistant

    A Russian-language voice assistant that works offline by default and supports extensible plugins and LLM-powered natural language command processing.

  • Project
    RchGrav/claudebox

    ClaudeBox is a Docker‑based tool that launches Anthropic’s Claude Code AI coding assistant inside an isolated container. It provides per‑project Docker images, persistent authentication/history, and a set of pre‑configured development profiles (C/C++, Python, Rust, Go, etc.). Features include multi‑instance support, firewall allowlists, macOS clipboard bridging, tmux integration, and a simple task engine. Installation is via a self‑extracting `.run` installer that also sets up Docker if needed. Users create “slots” (persistent Claude sessions) per project, add profiles, and run Claude with custom flags, all while keeping project data isolated and reproducible.

  • Project
    Tencent-Hunyuan/AuK

    AuK is a 1.5 B‑parameter open‑source foundation model for speech generation, editing, enhancement, and source separation. It works via natural‑language instructions, offers a high‑quality base model and a fast 4‑step distilled “Flash” version, and provides CLI, Gradio UI, Python API, fine‑tuning scripts, and ComfyUI integration.

  • Project
    Spielewoy/autoprompt-skill

    Autoprompt‑Skill is a Node‑based CLI that wraps multiple LLM coding agents (Claude, Codex, OpenCode, etc.) to automate the write‑test‑review loop. Install via npm, pick a provider, and run `autoprompt activate PROVIDER -- "<goal>"` (optionally with `path=` and concurrency flags). Benchmarks on Terminal‑Bench 2.1 show a 45 % drop in failures (73/89 solves vs. 60/89) at the cost of higher runtime and token usage.