---

Overview

Voice-Agents is a production-ready Python library for building enterprise-grade voice-enabled agentic applications. Built by Swarms Corporation, it provides seamless integration with multiple TTS/STT providers including OpenAI, ElevenLabs, and Groq, with real-time streaming capabilities optimized for agent-based architectures.

Voice-Agents delivers the infrastructure required to build conversational agentic assistants, voice-enabled agents, and real-time audio processing systems, enabling rapid deployment from prototype to production.

Built by Swarms Corporation

Voice-Agents is part of the Swarms ecosystem—the enterprise-grade, production-ready multi-agent orchestration framework. Learn more at swarms.ai and docs.swarms.world.

---

Features

Core Capabilities

Feature	Description
Multi-Provider TTS Support	Seamlessly switch between OpenAI, ElevenLabs, and Groq
Real-Time Streaming	Low-latency audio streaming for live agent interactions
Speech-to-Text	High-accuracy transcription using OpenAI Whisper
Audio Processing	Built-in utilities for recording, playback, and format conversion
Production-Ready	Enterprise-grade error handling, authentication, and logging

Advanced Features

Feature	Description
Streaming Callbacks	Real-time TTS callbacks for agent streaming outputs
Multiple Audio Formats	Support for PCM, MP3, Opus, AAC, FLAC, and more
Voice Customization	10+ OpenAI voices and 30+ ElevenLabs voices
Sentence Detection	Intelligent text formatting for natural speech pauses
FastAPI Integration	Generator-based streaming for web applications
Type Safety	Full type hints and Literal types for better IDE support

---

Installation

Basic Installation

pip install voice-agents

Development Installation

git clone https://github.com/The-Swarm-Corporation/Voice-Agents.git
cd Voice-Agents
pip install -e .

Requirements

• Python 3.10+
• API keys for your chosen providers:
  • OpenAI API key (for TTS and Whisper STT)
  • ElevenLabs API key (optional, for ElevenLabs TTS)

---

Quick Start

Environment Setup

Create a .env file or set environment variables:

export OPENAI_API_KEY="your-openai-api-key"
export ELEVENLABS_API_KEY="your-elevenlabs-api-key"  # Optional
export GROQ_API_KEY="your-groq-api-key"              # Optional

Basic Text-to-Speech

from voice_agents import stream_tts, format_text_for_speech

# Format text for natural speech
text = "Hello! This is a voice agent speaking. How can I help you today?"
chunks = format_text_for_speech(text)

# Convert to speech and play
stream_tts(chunks, model="openai/tts-1", voice="alloy")

Speech-to-Text

from voice_agents import speech_to_text, record_audio

# Record audio from microphone
audio = record_audio(duration=5.0, sample_rate=16000)

# Transcribe to text
transcription = speech_to_text(audio_data=audio, sample_rate=16000)
print(f"Transcribed: {transcription}")

---

Core Functions

Text-to-Speech (OpenAI)

from voice_agents import stream_tts, format_text_for_speech, VOICES

# Available voices: alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer
text_chunks = format_text_for_speech("Your text here")

# Basic usage - plays audio
stream_tts(text_chunks, model="openai/tts-1", voice="nova")

# Streaming mode for real-time processing
stream_tts(
    text_chunks,
    model="openai/tts-1",
    voice="alloy",
    stream_mode=True,  # Process chunks as they arrive
    response_format="pcm"
)

# For FastAPI/web streaming
from fastapi.responses import StreamingResponse

def audio_endpoint():
    generator = stream_tts(
        text_chunks,
        voice="alloy",
        return_generator=True
    )
    return StreamingResponse(generator, media_type="audio/pcm")

Text-to-Speech (ElevenLabs)

from voice_agents import stream_tts_elevenlabs, ELEVENLABS_VOICE_NAMES

# Available voices: rachel, domi, bella, antoni, elli, josh, and 25+ more
print(f"Available voices: {ELEVENLABS_VOICE_NAMES}")

# Basic usage
stream_tts_elevenlabs(
    text_chunks,
    voice_id="rachel",  # Use friendly name or voice ID
    model_id="eleven_multilingual_v2",
    stability=0.5,
    similarity_boost=0.75
)

# High-quality streaming for web
generator = stream_tts_elevenlabs(
    text_chunks,
    voice_id="domi",
    output_format="mp3_44100_128",  # Recommended for web
    return_generator=True
)

Speech-to-Text

from voice_agents import speech_to_text, record_audio
import numpy as np

# From audio file
transcription = speech_to_text(
    audio_file_path="recording.wav",
    model="whisper-1",
    language="en",  # Optional: auto-detect if None
    response_format="text"
)

# From numpy array (recorded audio)
audio = record_audio(duration=5.0, sample_rate=16000)
transcription = speech_to_text(
    audio_data=audio,
    sample_rate=16000,
    prompt="This is a technical conversation about AI"  # Optional context
)

# Get structured output
result = speech_to_text(
    audio_file_path="meeting.mp3",
    response_format="verbose_json"  # Returns detailed metadata
)

Audio Recording

from voice_agents import record_audio

# Record 5 seconds of audio
audio = record_audio(duration=5.0, sample_rate=16000, channels=1)

# Use with speech-to-text
from voice_agents import speech_to_text
text = speech_to_text(audio_data=audio, sample_rate=16000)

Streaming TTS Callback for Agents

from voice_agents import StreamingTTSCallback

# Create callback for real-time agent responses
tts_callback = StreamingTTSCallback(
    voice="alloy",
    model="openai/tts-1",
    min_sentence_length=10  # Minimum chars before speaking
)

# Use with any streaming text generator
def agent_stream():
    for chunk in some_agent.generate():
        tts_callback(chunk)  # Automatically speaks complete sentences
    tts_callback.flush()  # Speak any remaining text

Audio Format Utilities

from voice_agents import get_media_type_for_format

# Get MIME type for FastAPI
media_type = get_media_type_for_format("mp3_44100_128")
# Returns: "audio/mpeg"

media_type = get_media_type_for_format("pcm_44100")
# Returns: "audio/pcm"

---

Swarms Integration

Voice-Agents is designed to work seamlessly with Swarms, the enterprise-grade multi-agent orchestration framework.

Complete Example: Voice-Enabled Trading Agent

from swarms import Agent
from voice_agents import StreamingTTSCallback, format_text_for_speech

# Initialize the Swarms agent
agent = Agent(
    agent_name="Quantitative-Trading-Agent",
    agent_description="Advanced quantitative trading and algorithmic analysis agent",
    model_name="gpt-4",
    dynamic_temperature_enabled=True,
    max_loops=1,
    dynamic_context_window=True,
    top_p=None,
    streaming_on=True,
    interactive=False,
)

# Create the streaming TTS callback
tts_callback = StreamingTTSCallback(voice="alloy", model="openai/tts-1")

# Run the agent with streaming TTS callback
out = agent.run(
    task="What are the top five best energy stocks across nuclear, solar, gas, and other energy sources?",
    streaming_callback=tts_callback,
)

# Flush any remaining text in the buffer
tts_callback.flush()

print(out)

---

Examples

The examples/ directory contains comprehensive examples demonstrating all features of Voice-Agents, organized into logical categories. See the Examples README for detailed documentation.

Examples by Category

Text-to-Speech (examples/text_to_speech/)

Example File	Description
example_stream_tts.py	Unified TTS with OpenAI models, list_models()
example_stream_tts_elevenlabs.py	ElevenLabs TTS, unified and direct functions
example_streaming_tts_callback.py	StreamingTTSCallback for real-time TTS
example_voice_selection.py	Voice selection with list_voices()

Speech-to-Text (examples/speech_to_text/)

Example File	Description
example_speech_to_text.py	OpenAI Whisper transcription
example_speech_to_text_elevenlabs_file.py	ElevenLabs STT from audio file
example_speech_to_text_elevenlabs_audio.py	ElevenLabs STT from audio data

Utilities (examples/utilities/)

Example File	Description
example_format_text_for_speech.py	Text formatting for speech with abbreviation handling
example_play_audio.py	Audio playback and tone generation
example_record_audio.py	Microphone audio recording
example_get_media_type.py	Media type (MIME) utilities for FastAPI

Workflows (examples/workflows/)

Example File	Description
example_complete_voice_agent.py	Complete voice agent workflows

Running Examples

# Text-to-Speech examples
python examples/text_to_speech/example_stream_tts.py
python examples/text_to_speech/example_stream_tts_elevenlabs.py

# Speech-to-Text examples
python examples/speech_to_text/example_speech_to_text.py
python examples/speech_to_text/example_speech_to_text_elevenlabs_file.py

# Utility examples
python examples/utilities/example_format_text_for_speech.py
python examples/utilities/example_record_audio.py

# Workflow examples
python examples/workflows/example_complete_voice_agent.py

For more details, see the Examples README.

---

API Reference

Constants

• SAMPLE_RATE: Default sample rate (24000 Hz)
• VOICES: List of available OpenAI voices
• ELEVENLABS_VOICES: Dictionary mapping friendly names to ElevenLabs voice IDs
• ELEVENLABS_VOICE_NAMES: List of available ElevenLabs voice names
• OPENAI_TTS_MODELS: List of available OpenAI TTS models
• ELEVENLABS_TTS_MODELS: List of available ElevenLabs TTS models
• VoiceType: Type alias for OpenAI voice options

Functions

format_text_for_speech(text: str) -> List[str]

Intelligently formats text into speech-friendly chunks by detecting sentence boundaries, handling abbreviations, and preserving natural pauses.

stream_tts(text_chunks, model, voice, stream_mode, response_format, return_generator)

Unified TTS function supporting both OpenAI and ElevenLabs. Model format: "provider/model_name" (e.g., "openai/tts-1", "elevenlabs/eleven_multilingual_v2"). Returns generator for web streaming or plays audio directly.

list_models() -> List[dict]

List all available TTS models with their providers. Returns list of dictionaries with model, provider, and model_name keys.

list_voices() -> List[dict]

List all available voices from all providers. Returns list of dictionaries with voice, provider, voice_id, and description keys.

stream_tts_openai(text_chunks, voice, model, stream_mode, response_format, return_generator)

OpenAI TTS with streaming support. Returns generator for web streaming or plays audio directly.

stream_tts_elevenlabs(text_chunks, voice_id, model_id, stability, similarity_boost, output_format, return_generator)

ElevenLabs TTS with advanced voice control and multiple output formats.

speech_to_text(audio_file_path, audio_data, sample_rate, model, language, prompt, response_format)

OpenAI Whisper transcription with support for files or numpy arrays.

speech_to_text_elevenlabs(audio_file_path, audio_data, sample_rate, realtime, model_id, ...)

ElevenLabs Speech-to-Text with support for both real-time (WebSocket) and non-real-time (file upload) modes. Supports speaker diarization, timestamps, and language detection.

record_audio(duration, sample_rate, channels) -> np.ndarray

Record audio from default microphone. Returns numpy array.

play_audio(audio_data: np.ndarray)

Play audio data using sounddevice.

get_media_type_for_format(output_format: str) -> str

Get MIME type for audio format (useful for FastAPI).

Classes

StreamingTTSCallback

Real-time TTS callback for agent streaming outputs. Automatically detects complete sentences and converts them to speech.

Methods:

• __call__(chunk: str): Process streaming text chunk
• flush(): Speak any remaining buffered text

---

Use Cases

Conversational AI Assistants

Build voice-enabled chatbots and virtual assistants with natural, real-time speech synthesis.

Agent Narration

Provide audio feedback for long-running agent tasks, making agent behavior transparent and engaging.

Voice-Enabled Analytics

Create voice interfaces for data analysis, trading systems, and business intelligence tools.

Real-Time Transcription

Transcribe meetings, interviews, and conversations with high accuracy using Whisper.

Multi-Modal Applications

Combine voice input/output with visual interfaces for rich, interactive experiences.

---

Configuration

Environment Variables

# Required for OpenAI TTS and STT
OPENAI_API_KEY=your-key-here

# Required for ElevenLabs TTS
ELEVENLABS_API_KEY=your-key-here

Voice Selection

OpenAI Voices:

• alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer

ElevenLabs Voices:

• Professional: rachel, nicole, grace
• Expressive: domi, elli, bella
• Deep: antoni, josh, clyde
• And 20+ more (see ELEVENLABS_VOICE_NAMES)

---

Contributing

Voice-Agents is built by the community, for the community. We welcome contributions!

How to Contribute

1. Fork the repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Make your changes and add tests
4. Commit your changes: git commit -m 'Add amazing feature'
5. Push to the branch: git push origin feature/amazing-feature
6. Open a Pull Request

Development Setup

git clone https://github.com/The-Swarm-Corporation/Voice-Agents.git
cd Voice-Agents
pip install -e ".[dev]"
pre-commit install

Code Standards

• Follow PEP 8 style guidelines
• Add type hints to all functions
• Include docstrings for all public APIs
• Write tests for new features

---

License

This project is licensed under the Apache License - see the LICENSE file for details.

---

Acknowledgments

• Built by Swarms Corporation
• Part of the Swarms ecosystem
• Powered by OpenAI, ElevenLabs, and Groq APIs

---

Support & Community

• Documentation: GitHub Repository
• Swarms Documentation: docs.swarms.world
• Swarms Community: Discord
• Issues: GitHub Issues

---