feishu-speaker_skill

Overview

This skill provides two-way voice messaging for Feishu, enabling speech-to-text reception and text-to-speech sending with local Whisper transcription and Edge-TTS synthesis. It automates reply mode based on incoming message type and supports multiple Chinese voices and speed adjustments. The tool is designed for local transcription privacy and flexible TTS configurations.

How this skill works

The skill listens for Feishu voice messages, converts OGG/Opus audio to text locally using OpenAI Whisper, and passes the text into your workflow. For outgoing messages it renders text to high-quality audio via Edge-TTS, optionally processes audio with FFmpeg to produce Feishu-compatible formats, and sends via the Feishu API. A configurable reply mode chooses voice or text responses automatically or by override.

When to use it

• Enable natural voice interaction with Feishu bots or assistants
• Automate voice notifications, reminders, or scheduled broadcasts
• Add local speech-to-text to protect privacy and avoid cloud transcription
• Provide accessible responses for users who prefer audio over text
• Switch TTS voice and speed for different contexts (formal vs casual)

Best practices

• Install Whisper, Edge-TTS and FFmpeg on the host and verify model downloads before deployment
• Use reply_mode="auto" for seamless user experience, override to force voice/text as needed
• Limit single audio messages to recommended size (under 10 MB) to avoid upload errors
• Test chosen TTS voice and rate on sample phrases to find the most natural settings
• Keep Feishu App Secret and receiver IDs in the designated credentials file and restrict file permissions

Example use cases

• AI assistant: transcribe a user voice message, process intent, and reply by voice with current info
• Scheduled morning briefing: cron job runs a text announcement converted to audio and sent to a team
• Customer support: use a warm female TTS voice for polite, human-like replies to users
• Meeting reminders: send formal voice alerts with a deeper TTS voice 10 minutes before start
• Multi-tone announcements: switch voices and speeds for different announcement types