Orpheus Text to Speech

Generate expressive, natural-sounding speech with vocal direction controls for dynamic audio output.

Overview

Orpheus text-to-speech models by Canopy Labs provide fast, high-quality audio generation with unique expressive capabilities. Both models offer multiple voices and low-latency inference, with the English model supporting vocal direction controls for expressive performances.

Supported Models

Groq hosts two specialized Orpheus models for different language needs:

Pricing

API Endpoint

Quick Start

The speech endpoint accepts these parameters:

Basic Usage

# Install the Groq SDK:
# pip install groq

# English Model Example:
import os
from groq import Groq

client = Groq(api_key=os.environ.get("GROQ_API_KEY"))

speech_file_path = "orpheus-english.wav" 
model = "canopylabs/orpheus-v1-english"
voice = "troy"
text = "Welcome to Orpheus text-to-speech. [cheerful] This is an example of high-quality English audio generation with vocal directions support."
response_format = "wav"

response = client.audio.speech.create(
    model=model,
    voice=voice,
    input=text,
    response_format=response_format
)

response.write_to_file(speech_file_path)

// Install the Groq SDK:
// npm install --save groq-sdk

// English Model Example:
import fs from "fs";
import Groq from 'groq-sdk';

const groq = new Groq({
  apiKey: process.env.GROQ_API_KEY
});

const speechFilePath = "orpheus-english.wav";
const model = "canopylabs/orpheus-v1-english";
const voice = "hannah";
const text = "Welcome to Orpheus text-to-speech. [cheerful] This is an example of high-quality English audio generation with vocal directions support.";
const responseFormat = "wav";

async function main() {
  const response = await groq.audio.speech.create({
    model: model,
    voice: voice,
    input: text,
    response_format: responseFormat
  });
  
  const buffer = Buffer.from(await response.arrayBuffer());
  await fs.promises.writeFile(speechFilePath, buffer);
  
  console.log(`Orpheus English speech generated: ${speechFilePath}`);
}

main().catch((error) => {
  console.error('Error generating speech:', error);
});

curl https://api.groq.com/openai/v1/audio/speech \
  -X POST \
  -H "Authorization: Bearer $GROQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "canopylabs/orpheus-v1-english",
    "input": "Welcome to Orpheus text-to-speech. [cheerful] This is an example of high-quality English audio generation with vocal directions support.",
    "voice": "austin",
    "response_format": "wav"
  }' \
  --output orpheus-english.wav

import os
from groq import Groq

client = Groq(api_key=os.environ.get("GROQ_API_KEY"))

speech_file_path = "orpheus-arabic.wav" 
model = "canopylabs/orpheus-arabic-saudi"
voice = "fahad"
text = "مرحبا بكم في نموذج أورفيوس للتحويل من النص إلى الكلام. هذا مثال على جودة الصوت العربية السعودية الطبيعية."
response_format = "wav"

response = client.audio.speech.create(
    model=model,
    voice=voice,
    input=text,
    response_format=response_format
)

response.write_to_file(speech_file_path)

import fs from "fs";
import Groq from 'groq-sdk';

const groq = new Groq({
  apiKey: process.env.GROQ_API_KEY
});

const speechFilePath = "orpheus-arabic.wav";
const model = "canopylabs/orpheus-arabic-saudi";
const voice = "lulwa";
const text = "مرحبا بكم في نموذج أورفيوس للتحويل من النص إلى الكلام. هذا مثال على جودة الصوت العربية السعودية الطبيعية.";
const responseFormat = "wav";

async function main() {
  const response = await groq.audio.speech.create({
    model: model,
    voice: voice,
    input: text,
    response_format: responseFormat
  });
  
  const buffer = Buffer.from(await response.arrayBuffer());
  await fs.promises.writeFile(speechFilePath, buffer);
  
  console.log(`Orpheus Arabic speech generated: ${speechFilePath}`);
}

main().catch((error) => {
  console.error('Error generating Arabic speech:', error);
});

curl https://api.groq.com/openai/v1/audio/speech \
  -X POST \
  -H "Authorization: Bearer $GROQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "canopylabs/orpheus-arabic-saudi",
    "input": "مرحبا بكم في نموذج أورفيوس للتحويل من النص إلى الكلام. هذا مثال على جودة الصوت العربية السعودية الطبيعية.",
    "voice": "noura",
    "response_format": "wav"
  }' \
  --output orpheus-arabic.wav

Vocal Directions

Orpheus V1 English supports vocal directions using bracketed text like [cheerful] or [whisper] to control how the model speaks. This powerful feature enables everything from subtle conversational nuances to highly expressive character performances.

How Directions Work

• More directions = more expressive, acted performance
• Fewer/no directions = natural, casual conversational cadence
• Use 1-2 word directions (typically adjectives or adverbs)

Common use cases:

• Customer support: Use no directions for natural, friendly conversations
• Game characters: Add expressive directions for dynamic, performative speech
• Professional narration: Use [professionally] or [authoritatively] for business content
• Storytelling: Combine multiple directions to create engaging narrative performances

Direction Examples

Conversational tones:

• [cheerful], [friendly], [casual], [warm]

Professional styles:

• [professionally], [authoritatively], [formally], [confidently]

Expressive performance:

• [whisper], [excited], [dramatic], [deadpan], [sarcastic]

Vocal qualities:

• [gravelly whisper], [rapid babbling], [singsong], [breathy]

Note: There isn't an official or exhaustive list of directions; the model recognizes many natural descriptors and ignores vague or unfamiliar ones.

Using Vocal Directions

Available Voices

English Voices

The English model includes six professionally-trained voice personas. Each voice has different strengths for expressive direction performance.

Note: Some voices perform better with expressive directions than others. Experiment to find the voice that works best for your use case.

Arabic Saudi Dialect Voices

The Arabic model offers four distinct Saudi dialect voices with authentic pronunciation and regional nuances:

Use Cases

Best Practices

Punctuation control: Experiment with removing punctuation to give the model more freedom in choosing intonation patterns, especially for expressive performances.

Voice selection: Test different voices for your use case; some handle expressive directions better than others, particularly for complex emotional ranges.

Arabic considerations: Use proper Arabic script with diacritical marks. Test pronunciation with sample content before production deployment.

Limitations

Input length: The input text length is limited to 200 characters.

Batch processing: The batch processing API is not supported at this time for Orpheus models.