Clone a voice from a short reference audio sample and generate new speech in that voice. Useful for personalized TTS, localization, audiobook narration, and character voices.
Available models
| Model | Languages | Strengths |
|---|
| HiggsAudio V2.5 | 20+ | Highest fidelity, emotion preservation |
| Zonos v0.1 | 10+ | Fast inference, real-time apps |
| Chatterbox Multilingual | 30+ | Widest language coverage, cross-lingual |
Basic voice cloning
Provide a 10-30 second audio sample of the target voice:
from runcrate import Runcrate
client = Runcrate(api_key="rc_live_YOUR_API_KEY")
audio = client.models.text_to_speech(
model="bosonai/HiggsAudioV2.5",
input="Welcome to our quarterly earnings call. Strong growth across all segments.",
reference_audio="./ceo-sample.mp3",
)
with open("cloned-speech.mp3", "wb") as f:
f.write(audio)
import Runcrate from '@runcrate/sdk';
import { writeFileSync } from 'fs';
const rc = new Runcrate({ apiKey: 'rc_live_YOUR_API_KEY' });
const audio = await rc.models.textToSpeech({
model: 'bosonai/HiggsAudioV2.5',
input: 'Welcome to our quarterly earnings call.',
referenceAudio: './ceo-sample.mp3',
});
writeFileSync('cloned-speech.mp3', Buffer.from(audio));
curl https://api.runcrate.ai/v1/audio/speech \
-H "Authorization: Bearer rc_live_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "bosonai/HiggsAudioV2.5", "input": "Welcome to our quarterly earnings call.", "reference_audio": "https://example.com/ceo-sample.mp3"}' \
--output cloned-speech.mp3
reference_audio accepts a file path (auto base64-encoded), a URL, or raw base64 data.
All examples below reuse the same client.
Cross-lingual cloning
audio = client.models.text_to_speech(
model="ResembleAI/chatterbox-multilingual",
input="Bienvenidos a nuestra presentacion trimestral.",
reference_audio="./english-speaker.mp3", language="es",
)
with open("spanish-clone.mp3", "wb") as f:
f.write(audio)
Real-time cloning (Zonos)
audio = client.models.text_to_speech(
model="Zyphra/Zonos-v0.1-transformer",
input="Your order has been confirmed and will arrive by Thursday.",
reference_audio="./brand-voice.mp3",
)
with open("notification.mp3", "wb") as f:
f.write(audio)
Batch narration
chapters = [
{"file": "ch01.mp3", "text": "Chapter one. The morning light crept through the curtains."},
{"file": "ch02.mp3", "text": "Chapter two. The letter arrived on a Tuesday."},
{"file": "ch03.mp3", "text": "Chapter three. Three weeks had passed since the call."},
]
for ch in chapters:
audio = client.models.text_to_speech(
model="bosonai/HiggsAudioV2.5", input=ch["text"], reference_audio="./narrator.mp3",
)
with open(ch["file"], "wb") as f:
f.write(audio)
Tips
- Reference quality matters. Clean recording, minimal noise, 10-30s of clear speech.
- HiggsAudio for fidelity. When the clone must be indistinguishable from the original.
- Chatterbox for languages. 30+ languages, cross-lingual cloning.
