Skip to main content

Wit.ai TTS

Wit.ai provides text-to-speech capabilities as part of Facebook's AI services. It offers basic speech synthesis with support for multiple languages.

Authentication​

Wit.ai requires an API token for authentication:

from tts_wrapper import WitAiClient

tts = WitAiClient(credentials='your_wit_ai_token')
tip

Use environment variables for secure credential management:

import os

tts = WitAiClient(credentials=os.getenv('WITAI_TOKEN'))
note

WitAiClient can be used directly as it implements the TTS interface. The legacy pattern with separate WitAiTTS class is still supported for backward compatibility.

Features​

Voice Selection​

Set language for synthesis:

# Set language
tts.set_voice("en", "en-US")  # Language code and optional dialect

# Different languages
tts.set_voice("fr", "fr-FR")  # French
tts.set_voice("es", "es-ES")  # Spanish

Basic Synthesis​

Simple text-to-speech conversion:

# Basic speech synthesis
tts.speak("Hello, this is a test of Wit.ai TTS")

# Different languages
tts.set_voice("fr", "fr-FR")
tts.speak("Bonjour, ceci est un test")

File Output​

Save synthesized speech to file:

# Save as MP3
tts.synth_to_file("Hello world", "output.mp3", "mp3")

# Save as WAV
tts.synth_to_file("Hello world", "output.wav", "wav")

Best Practices​

  1. API Usage

    • Monitor API usage limits
    • Cache frequently used phrases
    • Handle rate limiting gracefully
    • Keep requests reasonable in frequency

  2. Performance

    • Reuse client instances
    • Handle network connectivity issues
    • Consider response times in your application

  3. Error Handling

    try:
        tts.speak("Hello, world!")
    except Exception as e:
        if "Unauthorized" in str(e):
            print("Check your Wit.ai token")
        elif "Connection" in str(e):
            print("Check internet connection")
        else:
            print(f"Error: {e}")

Limitations​

  • Requires internet connection
  • No SSML support
  • Limited voice selection
  • No word timing support
  • Basic prosody control
  • API rate limits apply
  • Quality may vary by language
  • Limited control over voice properties

Audio Settings​

Audio Format​

  • Format: MP3 (converted to WAV for playback)
  • Sample Rate: 22050 Hz
  • Channels: Mono (1 channel)
  • Sample Width: 16-bit
print(f"Audio rate: {tts.audio_rate}")    # 22050
print(f"Channels: {tts.channels}")        # 1
print(f"Sample width: {tts.sample_width}") # 2 (16-bit)

Language Support​

Wit.ai supports multiple languages:

# Common language codes
languages = {
    "English": ("en", "en-US"),
    "French": ("fr", "fr-FR"),
    "Spanish": ("es", "es-ES"),
    "German": ("de", "de-DE"),
    "Italian": ("it", "it-IT")
}

# Test different languages
for name, (lang, dialect) in languages.items():
    tts.set_voice(lang, dialect)
    tts.speak(f"This is a test in {name}")

Use Cases​

Basic TTS Applications​

# Simple announcements
tts.speak("Your download is complete")

# Multi-language messages
messages = {
    ("en", "en-US"): "Welcome to our service",
    ("fr", "fr-FR"): "Bienvenue à notre service",
    ("es", "es-ES"): "Bienvenido a nuestro servicio"
}

for (lang, dialect), message in messages.items():
    tts.set_voice(lang, dialect)
    tts.speak(message)

Integration with Wit.ai NLP​

While the TTS wrapper focuses on speech synthesis, Wit.ai also offers natural language processing capabilities that can be integrated:

# Example of combining NLP and TTS (requires separate Wit.ai NLP setup)
response = "Response from Wit.ai NLP"  # Your NLP logic here
tts.speak(response)

Additional Resources​

Next Steps​