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

Espresso

Espresso is an open-source, modular Python toolkit for training and decoding end-to-end neural automatic speech recognition models, built on PyTorch and fairseq.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Espresso — 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 Espresso when you need a local toolkit and example recipes for developing or experimenting with neural ASR models, rather than a hosted transcription application.

Who it is for

  • ASR researchers and developers working with PyTorch-based speech recognition.
  • Teams experimenting with distributed ASR training or speech-decoding approaches.

Documented uses

  • Train ASR models using recipes for WSJ, LibriSpeech or Switchboard.
  • Run distributed model training across GPUs and computing nodes.
  • Explore decoding methods, including look-ahead word-based language-model fusion.
  • Work with documented CTC, Transducer and Conformer options.
Before you choose

Requirements & limits

The repository describes Espresso as open-source and MIT-licensed. Local installation is documented; no hosted service or pricing details are provided.

  • Requires Python 3.8 or later and PyTorch 1.10 or later.
  • Training new models requires an NVIDIA GPU and NCCL.
  • The extra dependencies depend on the workflow: torchaudio for documented Transducer training; PyChain and OpenFst for LF-MMI; Kaldi for specified data and hybrid-system tasks.
  • The toolkit is built on PyTorch and fairseq and is configured for local development from source.

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

Install Espresso from source

Set up Espresso for local development with its documented core dependencies and Kaldi tools.

Before you start

  • Python 3.8 or later
  • PyTorch 1.10.0 or later
  • A compiled Kaldi directory

Steps

  1. Clone the Espresso repository and enter its directory: the command below then cd espresso.

    Documented command
    git clone https://github.com/freewym/espresso

  2. Install Espresso in editable mode with the command below.

    Documented command
    pip install --editable .

  3. Install kaldi_io, sentencepiece, and soundfile with the command below.

    Documented command
    pip install kaldi_io sentencepiece soundfile

  4. Build the tools with cd espresso/tools; make KALDI=.

  5. Add your Python path to PATH in examples/asr_/path.sh. The documented default is ~/anaconda3/bin.

Expected outcome

Espresso is installed in editable mode, with the listed Python packages and tools prepared for local development.

Guide limitations

  • Training new models additionally requires an NVIDIA GPU and NCCL.
  • Kaldi is needed for data preparation and feature extraction, scoring for some datasets such as Switchboard, and decoding for all hybrid systems.
  • The guide documents installation and tool setup, not running a training or decoding recipe.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
15 October 2026
Original website title
freewym/espresso · Official homepage title · retained in its source language
How we review sourcesSuggest a correction