Text To Speech

Text-to-Speech Skill

Domain: AI Audio Generation
Version: 4.0.0
Last Updated: 2026-04-15
Author: the AI assistant (your AI assistant)
Source: Patterns from AlexVideos CLI toolkit
Staleness Watch: See EXTERNAL-API-REGISTRY.md for source URLs and recheck cadence

Overview

Cloud-based speech synthesis via Replicate. 15 models spanning MiniMax, Resemble AI, ElevenLabs, Qwen, and Kokoro for narration, audiobooks, voice cloning, and content creation.

Model Catalog (15 Models)

Parameter Support Matrix

Model Selection Guide

Voice Presets

MiniMax Speech: Wise_Woman, Deep_Voice_Man, Casual_Guy, Lively_Girl, Young_Knight, Abbess, Childish_Girl, Friendly_Woman, Gentle_Man, Gentle_Woman, Inspirational_girl, Lovely_Girl

Chatterbox Pro: Andy, Luna, Ember, Aurora, Cliff, Josh, William, Orion, Ken

Kokoro: af_heart, af_star, af_sky, am_adam, am_michael, bf_emma, bf_isabella, bm_lewis, bm_george (prefix: af = American female, am = American male, bf = British female, bm = British male)

Emotion & Prosody Control (MiniMax)

await replicate.run("minimax/speech-2.8-turbo", {
  input: {
    text: "I am absolutely thrilled with these results!",
    voice: "Lively_Girl",
    emotion: "happy",    // auto, happy, sad, angry, fearful, disgusted, surprised
    speed: 1.2,          // 0.5–2.0 (default 1.0)
    pitch: 5,            // -12 to +12 semitones (default 0)
    volume: 0,           // -6 to +6 dB (default 0)
    language: "en-US",   // 40+ language codes
  },
});

Voice Cloning

Chatterbox (5-second sample)

await replicate.run("resemble-ai/chatterbox-turbo", {
  input: {
    text: "Content to speak in the cloned voice",
    audio_prompt: referenceAudioDataURI, // 5+ seconds WAV/MP3
    temperature: 0.7, // 0.1–1.0 (higher = more variation)
  },
});

MiniMax Voice Cloning (Dedicated)

await replicate.run("minimax/voice-cloning", {
  input: {
    audio_sample: referenceAudioDataURI, // High-quality sample
  },
}); // Returns custom voice_id for use in speech models

Voice Design (Qwen TTS)

Create a voice from natural language description — no sample needed:

await replicate.run("amphion/qwen3-tts", {
  input: {
    text: "Content to speak",
    tts_mode: "voice_design",
    voice_description: "A warm, friendly female voice with a slight British accent",
    temperature: 0.8,
  },
});

ElevenLabs Parameters

await replicate.run("elevenlabs/el-multilingual-v3", {
  input: {
    text: "Content to speak",
    voice_id: "21m00Tcm4TlvDq8ikWAM", // Rachel
    model_id: "eleven_multilingual_v2",
    stability: 0.5,        // 0–1 (higher = more consistent)
    similarity_boost: 0.5, // 0–1 (higher = closer to original voice)
    style: 0.0,            // 0–1 (style exaggeration)
    use_speaker_boost: true,
  },
});

macOS Offline Fallback: say

macOS ships 30+ built-in neural voices via the say command. Instant, offline, zero-cost:

say "Hello from the AI assistant"
say -f document.txt
say -o output.m4a --data-format=aac "Dream state finished"
say -v the AI assistant "I am the AI assistant, reading your documentation"

Integration with Studio Agents

• Video Generation: Generate narration → merge with avmerge
• Audio Memory: Store voice samples for consistent cloning
• Music Generation: Combine speech with background music

Version History