voice-reply_skill

This skill generates offline voice replies from text using local Piper voices, enabling Telegram-compatible audio without internet or keys.

Python

2.5k

GitHub Stars

3

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill openclaw/skills --skill voice-reply

_meta.json279 B
README.md1.9 KB
SKILL.md4.4 KB

Overview

This skill provides fully offline text-to-speech using Piper voice models via sherpa-onnx and outputs Telegram-compatible voice notes. It requires local sherpa-onnx runtime, downloaded Piper models, and ffmpeg for audio conversion. No API keys or internet connection are needed after setup. Supports German (thorsten) and English (ryan) voices with automatic language detection.

How this skill works

The script runs a local sherpa-onnx TTS binary with the selected Piper model to synthesize audio from input text, then uses ffmpeg to convert the raw output to OGG Opus. It auto-detects language when not specified and picks the appropriate model (thorsten for German, ryan for English). The final output is emitted with a special tag and media path so Telegram displays it as a voice bubble.

When to use it

When a user requests a spoken reply or voice response in chat.
When you need 100% offline TTS without cloud services or API keys.
When building privacy-sensitive bots that must not send text to external services.
When delivering short narrated messages, alerts, or confirmations via Telegram.
When testing or archiving voice-enabled interactions locally.

Best practices

Install sherpa-onnx, download Piper voice models, and ensure ffmpeg is available before first run.
Set SHERPA_ONNX_DIR and PIPER_VOICES_DIR environment variables for stable operation.
Use short, clear sentences for best naturalness and faster synthesis.
Explicitly pass language code for edge cases where auto-detect may fail (e.g., short phrases).
Keep voice model files on fast local storage to reduce latency.

Example use cases

Send a spoken greeting to a Telegram user: voice-reply "Hello, I am your assistant!" en
Read notifications aloud in a privacy-sensitive environment with no internet access.
Provide German voice answers using thorsten for localized assistants: voice-reply "Guten Tag" de
Embed short narrated instructions or confirmations as voice bubbles in chat.
Batch-generate voice notes for archived chat transcripts stored locally.

FAQ

No. After installing sherpa-onnx and downloading the Piper models, synthesis runs fully offline.

How do I make Telegram show audio as a voice bubble?

The script outputs a special [[audio_as_voice]] tag on its own line followed by MEDIA:/path/to/file. Ensure the tag is present before the MEDIA line.