docs(#1186): Add comprehensive multi-speaker TTS guide

[moguangyu5-design]

Summary

Adds comprehensive documentation for native multi-speaker TTS generation to address Issue #1186.

Changes Made

1. Created Multi-Speaker Guide (docs/multi_speaker_guide.md)

Complete documentation covering:

Four Usage Methods:

• CLI with direct reference audio
• CLI with reference ID (stored references)
• Python API direct integration
• REST API HTTP calls

Key Documentation:

• Reference audio requirements (3-10 seconds, clean audio)
• Memory caching for performance
• Multi-speaker architecture explanation
• Troubleshooting common issues
• Advanced usage patterns

API Reference:

• Complete ServeTTSRequest schema
• ServeReferenceAudio schema
• All parameters documented

2. Key Features Documented

Voice Cloning:

• Single vs multiple reference samples
• Quality requirements and best practices
• Common issues and solutions

Prosody Control:

• Natural language control methods
• Why emotion tags don't work (links to docs: Add emotion tags investigation documentation #1180)
• Alternative approaches

Advanced Usage:

• Multi-language code-switching
• Style transfer techniques
• Fine-grained parameter control

How Multi-Speaker Generation Works

1. Voice Encoding: Extract voice embedding from reference audio
2. Text Processing: Prepare text for synthesis
3. In-Context Learning: Model learns voice characteristics from references
4. Audio Generation: Synthesize speech in the target voice
5. Post-Processing: Format and return audio

Testing

Documentation includes working examples:

• All CLI commands tested
• API calls verified for syntax
• Python code is functional

Usage Example

python -m tools.api_client \
  --text "Hello, this is synthesized speech." \
  --reference_audio speaker.wav \
  --reference_text "This is the reference transcript." \
  --output output.wav

Related Issue

Resolves #1186

🤖 Generated with Claude Code