voicebox

Python text-to-speech library with built-in voice effects and support for multiple TTS engines.

| GitHub | Documentation 📘 | Audio Samples 🔉 |

# Example: Use gTTS with a vocoder effect to speak in a robotic voice

from voicebox import SimpleVoicebox
from voicebox.tts import gTTS
from voicebox.effects import Vocoder, Normalize

voicebox = SimpleVoicebox(
    tts=gTTS(),
    effects=[Vocoder.build(), Normalize()],
)

voicebox.say('Hello, world! How are you today?')

Setup

1. pip install voicebox-tts
2. Install the PortAudio library for audio playback.
   • On Debian/Ubuntu: sudo apt install libportaudio2
3. Install dependencies for whichever TTS engine(s) you want to use (see section below).

Supported Text-to-Speech Engines

Classes for supported TTS engines are located in the voicebox.tts package.

Amazon Polly 🌐

Online TTS engine from AWS.

• Class: voicebox.tts.AmazonPolly
• Setup: pip install "voicebox-tts[amazon-polly]"

ElevenLabs 🌐

Online TTS engine with very realistic voices and support for voice cloning.

• Class: voicebox.tts.ElevenLabsTTS
• Setup:
  1. pip install "voicebox-tts[elevenlabs]"
  2. Install ffmpeg or libav for pydub (docs)
  3. (Optional) Use an API key:

     from elevenlabs.client import ElevenLabs
     from voicebox.tts import ElevenLabsTTS
     
     tts = ElevenLabsTTS(client=ElevenLabs(api_key='your-api-key'))

eSpeak NG 🌐

Offline TTS engine with a good number of options.

• Class: voicebox.tts.ESpeakNG
• Setup:
  • On Debian/Ubuntu: sudo apt install espeak-ng

Google Cloud Text-to-Speech 🌐

Powerful online TTS engine offered by Google Cloud.

• Class: voicebox.tts.GoogleCloudTTS
• Setup: pip install "voicebox-tts[google-cloud-tts]"

gTTS 🌐

Online TTS engine used by Google Translate.

• Class: voicebox.tts.gTTS
• Setup:
  1. pip install "voicebox-tts[gtts]"
  2. Install ffmpeg or libav for pydub (docs)

🤗 Parler TTS 🌐

Offline TTS engine released by Hugging Face that uses a promptable deep learning model to generate speech.

• Class: voicebox.tts.ParlerTTS
• Setup: pip install git+https://github.com/huggingface/parler-tts.git

Pico TTS

Very basic offline TTS engine.

• Class: voicebox.tts.PicoTTS
• Setup:
  • On Debian/Ubuntu: sudo apt install libttspico-utils

pyttsx3 🌐

Offline TTS engine wrapper with support for the built-in TTS engines on Windows (SAPI5) and macOS (NSSpeechSynthesizer), as well as espeak on Linux. By default, it will use the most appropriate engine for your platform.

• Class: voicebox.tts.Pyttsx3TTS
• Setup:
  1. pip install "voicebox-tts[pyttsx3]"
  2. On Debian/Ubuntu: sudo apt install espeak

Effects

Built-in effect classes are located in the voicebox.effects package, and can be imported like:

from voicebox.effects import CoolEffect

Here is a non-exhaustive list of fun effects:

• Glitch creates a glitchy sound by randomly repeating small chunks of audio.
• RingMod can be used to create choppy, Doctor Who Dalek-like effects.
• Vocoder is useful for making monotone, robotic voices.

There is also support for all the awesome audio plugins in Spotify's pedalboard library using the special PedalboardEffect wrapper, e.g.:

from voicebox import SimpleVoicebox
from voicebox.effects import PedalboardEffect
import pedalboard

voicebox = SimpleVoicebox(
    effects=[
        PedalboardEffect(pedalboard.Reverb()),
        ...,
    ]
)

Examples

Minimal

# PicoTTS is used to say "Hello, world!"
from voicebox import SimpleVoicebox

voicebox = SimpleVoicebox()
voicebox.say('Hello, world!')

Pre-built

Some pre-built voiceboxes are available in the voicebox.examples package. They can be imported into your own code, and you can run them to demo:

# Voice of GLaDOS from the Portal video game series
python -m voicebox.examples.glados "optional message"

# Voice of the OOM-9 command battle droid from Star Wars: Episode I
python -m voicebox.examples.battle_droid "optional message"

Advanced

# Use eSpeak NG at 120 WPM and en-us voice as the TTS engine
from voicebox import reliable_tts
from voicebox.tts import ESpeakConfig, ESpeakNG, gTTS

# Wrap multiple TTSs in retries and caches
tts = reliable_tts(
    ttss=[
        # Prefer using online TTS first
        gTTS(),
        # Fall back to offline TTS if online TTS fails
        ESpeakNG(ESpeakConfig(speed=120, voice='en-us')),
    ],
)

# Add some voice effects
from voicebox.effects import Vocoder, Glitch, Normalize

effects = [
    Vocoder.build(),    # Make a robotic, monotone voice
    Glitch(),           # Randomly repeat small sections of audio
    Normalize(),        # Remove DC and make volume consistent
]

# Build audio sink
from voicebox.sinks import Distributor, SoundDevice, WaveFile

sink = Distributor([
    SoundDevice(),          # Send audio to playback device
    WaveFile('speech.wav'), # Save audio to speech.wav file
])

# Build the voicebox
from voicebox import ParallelVoicebox
from voicebox.voiceboxes.splitter import SimpleSentenceSplitter

# Parallel voicebox doesn't block the main thread
voicebox = ParallelVoicebox(
    tts,
    effects,
    sink,
    # Split text into sentences to reduce time to first speech
    text_splitter=SimpleSentenceSplitter(),
)

# Speak!
voicebox.say('Hello, world!')

# Wait for all audio to finish playing before exiting
voicebox.wait_until_done()

Command Line Demo

python -m voicebox -h               # Print command help
python -m voicebox "Hello, world!"  # Basic usage