UWebASR Client & Skill

A Python client for the UWebASR ASR service. This repository is structured as a Skill for agent, allowing AI agents to use it as a tool for speech-to-text tasks.

Usage

Quick Start (Standalone Script)

If you just want to use the transcription script directly without cloning the whole repository:

1. Download:

   curl -O https://raw.githubusercontent.com/honzas83/uwebasr-skill/main/scripts/uwebasr.py
   chmod +x uwebasr.py

2. Run:

   python3 uwebasr.py lindat/generic/cs/zipformer your_audio.mp3 --format txt

   Note: No external dependencies required. Only Python 3 and optionally ffmpeg for audio conversion.

General Command Structure

If you have cloned the repository, run from the root:

python3 scripts/uwebasr.py [INSTANCE/MODEL] [FILES...] --format [FORMAT]

Example:

python3 scripts/uwebasr.py lindat/generic/cs/zipformer HDS09.mp3 --format txt

Use the full UWebASR model path, including the instance prefix. Public UWebASR models listed on the service page use the lindat/ prefix, for example lindat/generic/cs/zipformer.

Supported Formats

• No --format: generate the default output set (speechcloud_json, txt, s.txt, vtt, s.vtt, json, trs, extended.trs).
• txt: Plain text
• s.txt: Segmented plain text using sp=0.3 and pau=2.0
• vtt: WebVTT subtitles
• s.vtt: Segmented WebVTT using sp=0.3 and pau=2.0
• json: Word-level API JSON with timestamps and confidence scores
• speechcloud_json: Raw SpeechCloud JSON
• trs: Transcriber XML
• extended.trs: Extended Transcriber XML with confidence values

Advanced Options

• --format: Generate only the selected format. Can be used multiple times.
• --n-workers: Number of parallel threads for processing multiple files.
• --no-ffmpeg: Skip local ffmpeg conversion and submit the file directly to the UWebASR API.
• --no-cookies: Disable cookie handling between API calls.
• --overwrite: Allow overwriting existing output files.
• --output-dir: Specify a custom directory for output files.
• --suffix: Add a custom suffix to output filenames.

Installation & Skill Setup

As a local repository

If you prefer to have the full project structure:

1. Clone the repository:

   git clone git@github.com:honzas83/uwebasr-skill.git
   cd uwebasr-skill

2. (Optional) External dependencies are no longer required, but you can check requirements.txt.

With Google Antigravity

Google Antigravity can use this repository as project context and run the client script from its terminal-enabled agent workflow.

1. Clone the skill into the project where you want to use it:

   cd /path/to/your/project
   git clone git@github.com:honzas83/uwebasr-skill.git

2. Add or update your project AGENTS.md with a short instruction that points Antigravity to this skill:

   Use the UWebASR skill from `uwebasr-skill/SKILL.md` for speech-to-text tasks.
   Run the client with `python3 uwebasr-skill/scripts/uwebasr.py [MODEL] [FILES...] --format [FORMAT]`.

3. In Antigravity, open the target project and ask the agent to transcribe an audio file using UWebASR.

As a Claude Code Skill

To install this skill in Claude Code:

1. Clone the repository into your personal Claude Code skills directory:

   git clone git@github.com:honzas83/uwebasr-skill.git ~/.claude/skills/uwebasr-client

2. Start or restart Claude Code. The skill is available as /uwebasr-client and can also be selected automatically when you ask for speech-to-text transcription.

For project-local installation, clone it into .claude/skills/uwebasr-client inside the target repository instead.

As a Codex Skill

To install this skill in Codex:

1. Clone the repository into your Codex skills directory:

   git clone git@github.com:honzas83/uwebasr-skill.git ~/.codex/skills/uwebasr-client

2. The skill will be automatically detected in your next Codex session.

License

MIT License. See LICENSE.

Acknowledgments & Citations

The UWebASR service is integrated into the national research infrastructure LINDAT/CLARIAH-CZ, which is part of the European CLARIN ERIC network.

Citing the service

If you use this service for your research, please cite the following papers:

• UWebASR Service and Zipformer Models: Švec, J., Lehečka, J., Ircing, P. (2025) Current State of the UWebASR - Web-Based ASR Service for Czech, Slovak, German, and English. CLARIN, ISSN 2773-2177.

• English, German, and Czech Models: Lehečka, J., Švec, J., Psutka, J.V., Ircing, P. (2023) Transformer-based Speech Recognition Models for Oral History Archives in English, German, and Czech. Proc. INTERSPEECH 2023, 201-205, doi: 10.21437/Interspeech.2023-872

• Slovak Model: Lehečka, J., Psutka, J.V., Psutka, J. (2023) Transfer Learning of Transformer-Based Speech Recognition Models from Czech to Slovak. In: Text, Speech, and Dialogue. TSD 2023. Lecture Notes in Computer Science, vol 14286. Springer, Cham. https://doi.org/10.1007/978-3-031-40498-6_29