Ttscn

ttscn — Multi-Platform Chinese TTS Skill

Overview

Generate natural speech audio from text. 15 backends — 11 China-friendly clouds plus 4 international or unified APIs (ElevenLabs / OpenAI / Google / Atlas Cloud).

New in 1.4–1.6: word-level timestamps (edge/azure/doubao/minimax/cosyvoice — best-effort, degrades to no boundaries), [PAUSE:x] + sound-tag markers (all platforms), --phonemes pronunciation overrides (azure/minimax). New in 1.7: --json flag (JSON envelope independent of --format), idempotency hits report cached: true and re-synthesize if the cached audio was deleted, --input f out.wav positional fixed, chunker never splits inside [PAUSE:x]. New in 1.8: MiniMax defaults to speech-2.8-hd (sound tags voiced out of the box), CosyVoice v3.5-flash supported (custom/cloned voices only — presets need cosyvoice-v3-flash). New in 1.9: Qwen3-TTS (DashScope, reuses DASHSCOPE_API_KEY), StepFun (reuses STEP_API_KEY), GLM-TTS (reuses ZHIPUAI_API_KEY) and Atlas Cloud TTS (asynchronous generation, bounded polling, credential-free media download, multilingual voice presets) backends — 15 backends total.

Cross-platform: Windows, macOS, Linux

All paths in this document are relative to this skill's root directory (the directory containing this SKILL.md) — resolve them against it.

When to Use This Skill

Automatically activate this skill when:

• User wants to convert Chinese text to speech audio
• Generating voice narration or voiceover for videos
• Creating audiobook or podcast audio from text
• User asks to compare TTS providers, choose a TTS backend, or see what voices are available
• User asks about TTS pricing, features, or which provider supports cloning/SSML/dialects
• User mentions any of: TTS, text-to-speech, 语音合成, 文字转语音, Edge TTS, Doubao TTS, CosyVoice, 火山引擎, 阿里云语音, Azure TTS, 腾讯云TTS, 百度语音, MiniMax, 讯飞语音, ElevenLabs, OpenAI TTS, Google Cloud TTS, Atlas Cloud TTS
• User needs word-level timestamps/subtitles, pause control, or fixing mispronounced Chinese characters (多音字)
• Any task where Chinese text-to-speech would be helpful

Provider Comparison Page

When the user wants to browse, compare, or choose a TTS provider, ALWAYS open the local HTML comparison page in their browser FIRST — it's a visual, filterable table that is much faster to scan than reading text output.

# Cross-platform (path relative to this skill's directory)
python3 -m webbrowser docs/providers.html

# …or the platform-native opener: open (macOS) / xdg-open (Linux) / start (Windows)

The comparison page includes:

• Filterable table — filter by free, SSML, voice cloning, streaming, dialects, multilingual
• Per-provider detail panels — cost, max chars/duration, clone method, emotion, languages
• Voice cards — recommended voices with style descriptions and best-use labels
• API key links — direct links to each provider's console for key acquisition

This page is auto-generated from data/providers.json (the single source of truth for all provider/voice data). Run python3 scripts/build_docs.py to regenerate it after editing the JSON. The same data is queryable on the CLI via python3 scripts/tts.py schema backends|voices (see Schema Introspection).

After opening the page, ask the user which backend and voice they'd like to use, then proceed to Step 2.

Workflow

Step 0 — Show the comparison page (when comparing/choosing)

If the user is browsing, comparing providers, or unsure which backend to use, open docs/providers.html as shown above. Let them explore, then ask which backend + voice they want.

Step 1 — Understand the request

Clarify what the user needs:

• Text: inline text or a file? Short or long-form?
• Voice style: male/female, young/mature, warm/energetic? (see voice guide below)
• Speed: normal, faster (+10-20%), slower (-10-20%)?
• Format: WAV (lossless) or MP3 (compressed)?

Step 2 — Pick a backend & voice

Choose based on the use case (see Backend Selection Guide). Default to Edge TTS with zh-CN-XiaoxiaoNeural (female, warm, standard) if unsure. Mention your choice.

Step 3 — Synthesize

Run scripts/tts.py with the text and chosen options.

Step 4 — Report

Confirm: output path, file size, audio duration.

Backend Selection Guide

Quick Pick

Full capability & voice data

The complete capability matrix (cost, max chars/duration per chunk, SSML, cloning method + cost, emotion, dialects, languages, streaming, setup difficulty) and the per-backend voice lists live in one place: data/providers.json. View them via:

• The comparison page (docs/providers.html) — best for humans
• python3 scripts/tts.py schema backends --full — full capability fields per backend
• python3 scripts/tts.py schema voices — every voice preset with style descriptions
• python3 scripts/tts.py --list — human-readable terminal summary

Do not maintain copies of these tables elsewhere — regenerate from the JSON.

Voice Cloning (clone command)

Create a custom voice from reference audio, store it under a name, then use