VoiceFuse

LiteLLM for Voice -- One API for every Voice AI provider.

English | Korean

Switch between OpenAI, ElevenLabs, Google Cloud TTS, and more --
by changing one parameter. No vendor lock-in. Automatic fallback. Built-in benchmarks.

---

Why VoiceFuse?

Every voice AI provider has a different API, different SDKs, different audio formats. Switching providers means rewriting code. Comparing them means maintaining parallel integrations.

VoiceFuse fixes this:

from voicefuse import VoiceFuse
vf = VoiceFuse()

audio = vf.tts("Hello world!", provider="openai", voice="alloy")

# Switch provider -- change one parameter. Voice is auto-mapped.
audio = vf.tts("Hello world!", provider="elevenlabs", voice="alloy")
# alloy (OpenAI) -> automatically mapped to Rachel (ElevenLabs)

Install

pip install voicefuse

# With audio playback support:
pip install voicefuse[play]

Quick Start

Python SDK

from voicefuse import VoiceFuse

vf = VoiceFuse()

# Basic TTS
audio = vf.tts("Hello!", provider="openai", voice="alloy")
audio.save("output.mp3")

# Auto fallback -- if primary fails, tries next
audio = vf.tts("Hello!", provider="elevenlabs", fallback=["openai", "google"])

# Cost-optimized -- picks cheapest provider
audio = vf.tts("Hello!", strategy="cheapest")

# Cross-provider voice -- use unified names
audio = vf.tts("Hello!", provider="openai", voice="female-warm")
audio = vf.tts("Hello!", provider="google", voice="female-warm")
# Both map to the best matching voice for each provider

CLI

# Generate and play speech
voicefuse tts "Hello world" -p openai -v alloy

# Save to file
voicefuse tts "Hello world" -p openai -v alloy -o output.mp3

# Compare providers side-by-side
voicefuse compare "Hello world" -p openai,elevenlabs,google

# Find matching voices across providers
voicefuse voices --match alloy

# Run full benchmark
voicefuse benchmark

# Check provider status
voicefuse providers

Provider Comparison

$ voicefuse compare "Hello world" -p openai,elevenlabs,google

  Provider Comparison
+------------+-------+---------+--------+--------+
| Provider   | Voice | Latency | Cost   | Size   |
+------------+-------+---------+--------+--------+
| google     | ...   | 180ms   | $0.0001| 9.8 KB |
| openai     | alloy | 340ms   | $0.0002| 12.4KB |
| elevenlabs | Rachel| 520ms   | $0.0003| 15.1KB |
+------------+-------+---------+--------+--------+

Cross-Provider Voice Matching

Use any voice name with any provider. VoiceFuse automatically maps to the closest match:

# "alloy" is an OpenAI voice, but works with any provider
audio = vf.tts("Hello", provider="elevenlabs", voice="alloy")
# -> maps to ElevenLabs "Rachel" (most similar)

# Or use universal voice names
audio = vf.tts("Hello", provider="openai", voice="female-warm")   # -> nova
audio = vf.tts("Hello", provider="elevenlabs", voice="female-warm") # -> Rachel
audio = vf.tts("Hello", provider="google", voice="female-warm")    # -> en-US-Neural2-C

Available unified voices: female-warm, female-bright, female-calm, male-deep, male-warm, male-narrative

Configuration

Set API keys via environment variables:

export OPENAI_API_KEY=sk-...
export ELEVENLABS_API_KEY=...
export GOOGLE_API_KEY=...

Or use voicefuse.yaml:

providers:
  openai:
    api_key: ${OPENAI_API_KEY}
  elevenlabs:
    api_key: ${ELEVENLABS_API_KEY}

defaults:
  provider: openai
  fallback: [elevenlabs, google]

Supported Providers

Provider	TTS	STT	Voice Cloning	Status
OpenAI	Yes	Soon	--	Available
ElevenLabs	Yes	--	Soon	Available
Google Cloud	Yes	Soon	--	Available
Deepgram	--	Soon	--	Coming Soon
Azure	Soon	Soon	--	Coming Soon
AWS Polly	Soon	--	--	Coming Soon

Contributing

Adding a provider is easy. See CONTRIBUTING.md.

git clone https://github.com/Daewooki/voicefuse.git
cd voicefuse
pip install -e ".[dev]"
pytest -v  # 70+ tests

License

Apache 2.0 -- see LICENSE