yangagent/minimax-tts-pipeline-skill
API developmentThis page was translated automatically. Technical details may differ from the original. View the original
A TTS Skill for generating high-quality audio from Chinese verbal transcripts via the MiniMax TTS API, automatically handling high-frequency pronunciation errors such as polyphonic characters, English abbreviations, brand names, and version numbers.
Overview
A TTS Skill for generating high-quality audio from Chinese verbal transcripts via the MiniMax TTS API, automatically handling high-frequency pronunciation errors such as polyphonic characters, English abbreviations, brand names, and version numbers. Each time you run the input file in the same directory, create tts-YYYYMMDD-HHMMSS/ directory: Install using the npx skills add YangAgent/minimax-tts-pipeline command Use MiniMax to generate spoken audio, with the content file path to /path/to/your/script.txt - Add: "Qwen is pronounced as Qianwen" - Query: "See what pronunciation rules are there" - Delete: "Delete Qwen's rules" - Modify: "Change Qwen's pronunciation to Qianwen San" Modify the voice_setting object in scripts/call_tts.py: - : Text replacement, tone generation, and API calls are all executed by Python scripts - : Only one terms.json is maintained throughout the process, and the same file is directly modified at each step - : After each stage is completed, a validation script is run; if it fails, it stops and does not proceed to the next stage - : MiniMax TTS itself can correctly read most common English words aloud, adding tone rules only for words that are truly prone to misreading - Python 3.10+ - requests library - MiniMax API Key (Speech 2.8 Turbo model)
Overview (machine translated)
A TTS Skill for generating high-quality audio from Chinese verbal transcripts via the MiniMax TTS API, automatically handling high-frequency pronunciation errors such as polyphonic characters, English abbreviations, brand names, and version numbers.
Workflow
Core Data Stream:
Output the file
Each time you run, create the tts-YYYYMMDD-HHMMSS/ directory in the same folder as the input file:
| Document | Instructions |
|---|---|
input.raw.txt |
Original input (read-only) |
input.txt |
Punctuation normalized input (read-only) |
terms.json |
A unique, structured work document for the entire process |
normalized.txt |
Standardized text |
output.wav |
TTS output audio |
output.title |
MiniMax returns font-level timestamp subtitles JSON |
output.srt |
SRT subtitle file |
Get Started Quickly
1. Install the skill
Install using the npx skills add YangAgent/minimax-tts-pipeline command
2. Configure the API Key
Configure in skills/minimax-tts-pipeline/.env:
When using it for the first time, Skill will automatically detect and prompt you to enter.
3. Usage
Trigger the skill in the Agent by providing the path to the text file:
Use MiniMax to generate spoken audio, with the content file path: /path/to/your/script.txt
4. Custom pronunciation rules
Supports managing pronunciation rules through conversations, with rules persisted to user-rules.json:
- Added:
"Qwen 都读作千问" - Query:
"看看有哪些发音规则" - Deleted:
"删掉 Qwen 的规则" - Modification:
"把 Qwen 的读法改成千问三"
User rules have the highest priority and are prioritized in each step of processing.
5. Adjust sound parameters
Modify the voice_setting object in scripts/call_tts.py:
| Parameter | Description | Range | Default value |
|---|---|---|---|
voice_id |
Tone ID | See MiniMax documentation | male-qn-qingse |
speed |
Speech Speed | [0.5, 2] |
1 |
vol |
Volume | (0, 10] |
1 |
pitch |
Intonation | [-12, 12] |
0 |
Project Structure
Design Principles
- LLM only changes terms.json: text replacement, tone generation, and API calls are all executed by Python scripts
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add yangagent/minimax-tts-pipeline-skill