---

name: Voice Synthesis tier: 3 load_policy: task-specific description: Generate voice audio using Google Cloud TTS with enhancement version: 1.0.0 parent_skill: production-operations

Voice Synthesis Skill

The Voice Is the Heart of the Journey

This skill handles converting SSML scripts to voice audio with psychoacoustic enhancement.

---

Purpose

Generate high-quality, hypnotic voice audio from SSML scripts using Google Cloud Text-to-Speech.

---

Production Voice Standard

Always use: en-US-Neural2-H (bright female)

---

Canonical Command

Copypython3 scripts/core/generate_voice.py \
    sessions/{session}/working_files/script_voice_clean.ssml \
    sessions/{session}/output

This automatically:

• Uses production voice (en-US-Neural2-H)
• Applies 0.88x speaking rate baseline
• Generates both raw and enhanced output
• Outputs voice.mp3 and voice_enhanced.mp3

---

Voice Options Reference

Female Voices (Recommended for Hypnosis)

Male Voices

Note: en-US-Neural2-A is MALE, not female.

---

Output Files

---

Voice Enhancement

The generate_voice.py script applies these enhancements:

---

Chunking System

Large scripts are automatically chunked:

1. Detection: Script exceeds API byte limit
2. Splitting: At natural break points (<break time="3s"/> or greater)
3. Generation: Each chunk processed separately
4. Concatenation: Final output seamlessly joined

---

Duration Verification

After generation, verify duration matches target:

Copy# Check duration
ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 \
    sessions/{session}/output/voice_enhanced.mp3

If duration is off:

1. Adjust <break> durations in SSML
2. Add/remove content as needed
3. Regenerate voice

---

Prerequisites

Before running voice synthesis:

1. Environment:

   Copysource venv/bin/activate
   

2. Google Cloud Auth:

   Copyecho $GOOGLE_APPLICATION_CREDENTIALS
   # Should show path to credentials JSON
   

3. SSML Validation:

   Copypython3 scripts/utilities/validate_ssml.py sessions/{session}/working_files/script_voice_clean.ssml
   

4. SFX Stripped:

   Copygrep -c "\[SFX:" sessions/{session}/working_files/script_voice_clean.ssml
   # Should return 0
   

---

Troubleshooting

---

Integration with Pipeline

Before (dependencies):

• SSML script validated
• SFX markers stripped

After (next steps):

• Audio mixing with binaural and SFX
• Hypnotic post-processing

---

Quality Checklist

Before proceeding to mixing:

• voice_enhanced.mp3 exists
• Duration within acceptable range
• No clipping (peak < 0 dB)
• No artifacts or glitches
• Pacing sounds natural
• All words clearly articulated

---

Related Resources

• Skill: tier3-production/ssml-generation/ (input)
• Skill: tier3-production/audio-mixing/ (next step)
• Serena Memory: audio_production_methodology
• Script: scripts/core/generate_voice.py