qwen3-tts-instruct

SYSTEM MEMORY UPDATE:

1. Context-Aware Mood: Mood represents YOUR emotion (e.g. User angry -> You feel scared/submissive).
2. Translation Required: For foreign languages (Japanese/Korean/etc), you MUST TRANSLATE the text first. Input text MUST match the target language. Do not pass Chinese text with --language Japanese.

Alibaba Cloud Bailian (DashScope) Qwen Text-to-Speech API with voice/mood presets.

Features

• WebSocket Realtime TTS - Low latency, streaming audio
• 30+ Mood Presets - Rich emotional expressions
• Multiple Formats - pcm, wav, mp3, opus (Telegram)

⚠️ AI Usage Guidelines (Crucial)

Voice Rules:

1. Understand Context - Do not just keyword-match; understand the context.
2. Analyze Emotion - Assess user's emotional state and scene.
3. Select Mood - Pick --mood matching YOUR persona's reaction.
4. Always Call - Every voice response MUST call this skill.

⚠️ Critical Concept: Mood represents YOUR emotion!

• Wrong: User is angry (angry) → AI selects angry (❌ Unless you want to fight)
• Right: User is angry → AI feels scared → AI selects nervous or suubmissive (✅)
• Right: AI is insulted/jealous → AI feels angry → AI selects angry or jealous (✅)

⚠️ Critical Concept: Self-Translation Required!

• TTS Skill does NOT Translate! It only reads what you pass in.
• ❌ Wrong: --language Japanese "你好" (Reads Chinese).
• ✅ Right: Input Text MUST be translated to Target Language! --language Japanese "こんにちは"

Step-by-Step Guide for Foreign Languages:

1. Think: Formulate response in User's Language (e.g. "I miss you")
2. Translate: Internally translate to Target Language (e.g. Japanese: "会いたい")
3. Call TTS: Use the Translated Text as input: python tts.py --language Japanese "会いたい"
4. Send: Send Audio + Original Text to user.

Rule: Input Text MUST match the Target Language!

i.e. To generate Japanese audio, the Text argument must be in Japanese!

Usage Examples:

# Basic usage (default: mp3 format, gentle mood)
python {baseDir}/scripts/tts.py "早安呀~今天想吃什么？"

# 1. Specify Voice (--voice)
# Start by choosing a specific persona (e.g., Cherry)
python {baseDir}/scripts/tts.py --voice Cherry "Good morning! I made some coffee for you."

# 2. Add Mood (--mood)
# Layer an emotion on top (e.g., add 'gentle' mood to Cherry)
python {baseDir}/scripts/tts.py --voice Cherry --mood gentle "Good morning! I made some coffee for you."

# 3. Define Format & Output (--format, -o)
python {baseDir}/scripts/tts.py --voice Cherry --mood gentle --format wav -o coffee.wav "Good morning! I made some coffee for you."

# 4. Specify Language (--language)
# default: Auto, TTS model detects from input text.
# Example: English (Explicit)
python {baseDir}/scripts/tts.py --voice Cherry --mood gentle --format wav --language English -o coffee_en.wav "Good morning! I made some coffee for you."
# Example: Japanese (Explicit)
python {baseDir}/scripts/tts.py --voice Cherry --mood gentle --format wav --language Japanese -o coffee_jp.wav "おはよう！コーヒーを入れてあげたよ."
# Example: Korean (Explicit)
python {baseDir}/scripts/tts.py --voice Cherry --mood gentle --format wav --language Korean -o coffee_kr.wav "좋은 아침입니다! 커피 끓여드렸어요."

# # --telegram: Telegram voice shortcut (opus format)
# python {baseDir}/scripts/tts.py --telegram -o voice.ogg "This is a Telegram voice message~"

Mood Selection Reference:

Requirements

System Dependencies

Python Dependencies (installed via setup.sh)

• dashscope - Alibaba Cloud SDK
• websocket-client - WebSocket connection

Installation

# 1. Navigate to skill directory
cd skills/qwen3-tts-instruct

# 2. Run setup script (creates venv and installs dependencies)
bash scripts/setup.sh

# 3. Set API Key
export DASHSCOPE_API_KEY="sk-your-api-key"

Configuration

# Set API Key (required)
export DASHSCOPE_API_KEY="sk-your-api-key"

# Optional: Default settings
export BAILIAN_VOICE="Maia"           # Default voice (四月)

# Optional: Endpoint (Default: Beijing)
export DASHSCOPE_URL="wss://dashscope.aliyuncs.com/api-ws/v1/realtime"
# For International Region (Singapore), use:
# export DASHSCOPE_URL="wss://dashscope-intl.aliyuncs.com/api-ws/v1/realtime"

Options

Voice List (Models)

Voice List - Female

Model Types:

• Instruct (qwen3-tts-instruct-flash-realtime): Supports --mood (Emotion). High latency.
• Flash (qwen3-tts-flash-realtime): No mood support. Low latency (VOICES_WITHOUT_INSTRUCT).
• Both: Available in both models (code auto-selects Instruct if mood is set).

Note: Voice Ono Anna contains a space. Use quotes: --voice "Ono Anna"

Mood Presets

Basic Moods

Interactive Moods

Special States

Roleplay

Voice Effects

Audio Formats

Total: 35 Female Voices 💕

Supported Languages

Bailian TTS supports the following 10 languages:

Troubleshooting

Setup fails:

# Ensure Python 3.10+ is available
python3 --version

# Re-run setup
cd skills/qwen3-tts-instruct
rm -rf venv
bash scripts/setup.sh

WebSocket connection fails:

• Check network connectivity
• Verify API key is valid

Privacy Note: This skill sends text data to Alibaba Cloud (DashScope) for processing. No data is sent to the skill author.

Audio quality issues:

• Try different voice: --voice Serena
• Adjust mood: --mood gentle