speech-to-text

What This Skill Does

The speech-to-text skill for OpenClaw provides high-fidelity audio transcription using state-of-the-art models like Fast Whisper Large V3 and Whisper V3 Large. By integrating the inference.sh CLI, the skill enables users to seamlessly convert spoken language into written text. It supports over 99 languages and offers advanced capabilities such as timestamp generation, language detection, and even automatic translation to English. This skill is engineered to handle diverse audio sources, including podcasts, voice notes, research interviews, and meeting recordings, providing accurate, structured data that is easy to search, index, or archive. By leveraging the inference.sh infrastructure, it delivers performance-optimized transcription directly within your workflow.

Installation

To integrate this skill into your environment, use the OpenClaw management CLI. Run the following command in your terminal:

clawhub install openclaw/skills/skills/okaris/speech-to-text

Ensure that you have the inference.sh CLI configured beforehand by running 'infsh login'. This ensures that the skill can authenticate against the inference engine to pull the necessary models and process your audio files efficiently.

Use Cases

• Corporate Meetings: Automatically generate searchable records of board meetings or team syncs.
• Content Creation: Effortlessly transcribe podcast episodes for show notes or blog posts.
• Accessibility: Provide accurate subtitles for video content to make audio media inclusive.
• Voice Productivity: Quickly convert dictated voice notes into organized, actionable text documentation.
• Multilingual Research: Translate and transcribe interviews conducted in foreign languages to streamline analysis.

Example Prompts

1. "Transcribe this meeting recording located at https://audio-source.mp3 and provide me with a full transcript including timestamps."
2. "Can you take this French interview from https://interview.mp3 and translate it into English for my report?"
3. "Generate a subtitle file for this video https://video.mp4 using the Whisper V3 Large model for maximum accuracy."

Tips & Limitations

• Model Selection: Choose 'Fast Whisper' for high-speed batch processing, or 'Whisper V3 Large' if your priority is the highest possible word-error-rate accuracy.
• Audio Quality: Ensure source audio is clear and free of extreme background noise; while Whisper is robust, high-quality input significantly improves punctuation and technical term accuracy.
• Network Dependence: Because this skill utilizes the inference.sh API, you will need a stable internet connection to upload your audio files and retrieve the generated text results.
• Formatting: The output is returned as structured JSON, making it ideal for piping into other OpenClaw workflows or automated database entries.