OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Efficient Speech Codec

Efficient Speech Codec is an open-source repository for ESC, a neural speech codec that compresses 16 kHz speech into a coded representation and reconstructs audio from it. It provides pretrained checkpoints, a command-line compression workflow, Python encoder and decoder examples, and resources for training and evaluating codec models.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Efficient Speech Codec — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use ESC when a local Python repository workflow suits speech compression, reconstruction, or codec experimentation, and you can provide a model checkpoint and compatible speech input.

Who it is for

  • Developers and researchers working on speech compression or neural audio codecs.
  • Users who want to encode 16 kHz speech and save reconstructed audio with a pretrained checkpoint.
  • Researchers adapting or evaluating codec models with their own speech data.

Documented uses

  • Compress 16 kHz speech at one of six documented bitrates, from 1.5 to 9 kbps.
  • Reconstruct audio from the compressed codes.
  • Use pretrained checkpoints as a starting point for codec experiments.
  • Train or evaluate models using speech datasets and the repository’s documented workflows.
Before you choose

Requirements & limits

The GitHub repository is public and identifies its license as MIT. The README documents a Python 3.8 Conda environment and dependency installation from requirements.txt.

  • The documented input focus is 16 kHz speech.
  • The num_streams setting ranges from 1 to 6, corresponding to 1.5–9 kbps.
  • The command-line workflow produces both a .pth code file and reconstructed .wav audio.
  • Pretrained ESC and DAC checkpoints are listed in the repository.
  • Custom training uses Accelerate and Weights & Biases; the README’s base-model example specifies four RTX 4090 GPUs.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Compress speech and save reconstructed audio

Encode an input WAV file and produce its compressed code and reconstructed WAV using a pretrained model.

Before you start

  • Python 3.8 in a Conda environment
  • Repository dependencies installed with pip install -r requirements.txt
  • An input WAV file and a downloaded model checkpoint
  • A save directory and a chosen num_streams value from 1 to 6

Steps

  1. Create and activate the esc Conda environment, then install the repository requirements.

  2. Download an ESC checkpoint from the repository’s model-checkpoints section.

  3. Run the documented scripts.compress command with input, save_path, model_path, num_streams and device arguments.

  4. Find the resulting .pth code and reconstructed .wav under the specified save_path.

Expected outcome

A compressed representation and reconstructed speech file; num_streams 1–6 corresponds to 1.5–9 kbps.

Guide limitations

  • The repository describes 16 kHz speech; it does not document a conversion workflow for other sample rates.
  • This recipe follows repository instructions and is not an execution test.
  • The reported comparative reconstruction quality is not a guarantee for a particular input.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
yzGuu830/efficient-speech-codec · Official homepage title · retained in its source language
How we review sourcesSuggest a correction