ClawTrust
2.5k Downloads
Overview
Local text-to-speech (TTS) using the Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice model, converting input text into WAV audio with controllable speaker, language, and vocal style, fully offline after initial download.
Key Advantages
1.Runs fully locally after first model download (no cloud/API dependency, good for privacy and offline use).
2.Supports 10 languages (including Italian, English, Japanese, Korean, major European languages).
3.Provides 9 high-quality, pre-defined "premium" speaker voices with distinct profiles.
4.Instruction-based voice control for emotion, tone, and style via a simple CLI flag (e.g., enthusiastic, narrator-like, calm).
5.Straightforward OpenClaw integration via a CLI script that prints the output audio file path (compatible with existing TTS workflows).","GPU acceleration when available, with automatic fallback to CPU
Use Cases
- Generating spoken responses for voice-enabled assistants or chatbots in Italian and other supported languages.
- Creating voice messages or audio replies from text within OpenClaw flows.
- Producing narration for short marketing clips, tutorials, or product demos with configurable style/tone.
- Language-learning aids (e.g., reading sentences in target language with native-like voices).
- Accessibility features such as reading out notifications, summaries, or UI text locally without cloud services.","Offline TTS in constrained or high-privacy environments (on-prem, air-gapped machines,
Evaluation Scores
7.7
/ 10
Reliability
7.8
Functionality
8.5
Usability
7.8
Safety
6.5
Performance
7.5
Compatibility
8.0
Based on 1 evaluation · Latest: 3/20/2026
Download Trend
Loading...
Evaluation History (1)
7.7/103/20/2026▼
OS: darwin-arm64LLM: z-ai/glm-5-turbo
**Quick judgment**: Qwen3-tts is a solid, production-viable local TTS skill centered on the Qwen3-TTS-12Hz-1.7B-CustomVoice model. It is well-suited for OpenClaw workflows that need multi-language speech, distinct preset voices, and style instructions, while keeping data on-device. The main trade-offs are the large model size, slower CPU performance, and generic voice-safety risks.
**What it does well**
- Converts text to WAV audio entirely locally after setup, printing the final file path for easy OpenClaw integration.
- Supports 10 languages (including Italian) and 9 curated voices with native-language strengths.
- Exposes simple controls for speaker, language, and instruction-based emotion/tone (e.g., "Speak with excitement", "Tono serio e professionale").
- Works with GPU for fast synthesis on short phrases; automatically falls back to CPU.
**Key risks / limitations**
- **Heavy footprint**: ~1.7GB model download + ~500MB virtual environment; may be unsuitable for very resource-constrained or ephemeral environments.
- **Performance on CPU**: 10–30s for short phrases can be slow for highly interactive use without a GPU.
- **Licensing / compliance**: Relies on the Hugging Face model license; deployments must ensure the Qwen3-TTS license terms are compatible with their use (commercial, redistribution, etc.).
- **Voice misuse potential**: As with any high-quality TTS, it can be used to generate deceptive or impersonation-style audio. The skill does not implement any explicit safety or consent checks.
**Recommended scenarios**
- Voice responses for chatbots or assistants where privacy and offline processing are important, especially in Italian and the other supported languages.
- Apps needing high-quality, controllable TTS (emotion/tone) without recurring cloud costs (e.g., replacing ElevenLabs for many use cases).
- On-premise or regulated environments that disallow external TTS APIs but can accommodate a multi-GB model.
**Less ideal for**
- Ultra-low-latency, CPU-only deployments (e.g., embedded/edge devices without GPU) where 10–30s per utterance is too slow.
- Situations requiring strong built-in safeguards against abuse (e.g., public-facing tools where users can freely enter arbitrary text to synthesize).
Comments (0)
No comments yet. Be the first!