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

Edgedict

Edgedict is a Python RNN-Transducer model and codebase for online speech recognition and speech-to-text. Its repository describes a model trained on more than 2,000 hours of speech, pretrained weights, and decoding workflows for live video and audio streams.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Edgedict — 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 Edgedict when you need a code-based starting point for incremental speech recognition from live video or streamed audio, and can work with its Python scripts, configuration files and pretrained weights.

Who it is for

  • Developers adding online speech recognition to applications that process spoken audio.
  • Teams prototyping live captions or streaming transcription with an RNN-T model.
  • Practitioners comparing PyTorch, ONNX and OpenVINO inference for this model.

Documented uses

  • Decode speech from a YouTube Live video into text.
  • Process audio through a streaming speech-recognition script, including a documented FLAC example.
  • Build a speech-to-text or live-captioning component around an online RNN-T model.
  • Explore model inference through ONNX- or OpenVINO-compatible tooling.
Before you choose

Requirements & limits

  • The repository provides Python scripts and example flag files for YouTube Live and streaming audio.
  • Pretrained weights are identified as available through a release, while project setup instructions are referenced in the README.
  • The repository reports ONNX and OpenVINO ports; its own evaluation found OpenVINO inference slower than PyTorch and ONNX Runtime in the specified test setup.
  • The README reports a 16.3% word-error rate for its best model on Librispeech test-clean, so target accuracy should be assessed for the intended data and use.
  • The README characterizes RNN-T training as resource-intensive; this training context is separate from the requirements for running pretrained weights.

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

Run the documented streaming-audio example

Pass an audio file to Edgedict’s streaming script for speech recognition.

Before you start

  • The Edgedict repository and its project setup completed according to its instructions.
  • The pretrained model weights and configuration referenced by the example.
  • An audio file in the example’s supported input format; the README demonstrates a FLAC path.

Steps

  1. Follow the repository’s project setup instructions.

  2. Obtain the pretrained weights from the linked release and use the model name and flag file shown in the README.

  3. Invoke stream.py with the documented flag file, model name and a --path argument pointing to the audio file.

Expected outcome

The documented workflow sends the specified audio file through the streaming speech-recognition script.

Guide limitations

  • The recipe does not establish accuracy or performance for a particular input or machine.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
15 October 2026
Original website title
theblackcat102/edgedict · Official homepage title · retained in its source language
How we review sourcesSuggest a correction