@mastra/voice-deepgram

Deepgram voice integration for Mastra, providing both Text-to-Speech (TTS) and Speech-to-Text (STT) using Deepgram's Aura (TTS) and Nova (STT) families.

Installation

npm install @mastra/voice-deepgram

Configuration

The module requires a Deepgram API key, which can be provided through environment variables or directly in the configuration:

DEEPGRAM_API_KEY=your_api_key

Usage

import { DeepgramVoice } from '@mastra/voice-deepgram';

// Create voice with both speech and listening capabilities
const voice = new DeepgramVoice({
  speechModel: {
    name: 'aura', // TTS family
    apiKey: 'your-api-key', // Optional, can use DEEPGRAM_API_KEY env var
  },
  listeningModel: {
    name: 'nova', // STT family
    apiKey: 'your-api-key', // Optional, can use DEEPGRAM_API_KEY env var
  },
  speaker: 'asteria-en', // default voiceId (see voice.ts)
});

// List available voices
const voices = await voice.getSpeakers();

// Generate speech
const audioStream = await voice.speak('Hello from Mastra!', {
  speaker: 'hera-en', // override speaker voice
});

// Convert speech to text
const result = await voice.listen(audioStream, {
  diarize: true,
  diarize_speaker_count: 2,
});
console.log(result.transcript);

Features

• High-quality Text-to-Speech synthesis
• Accurate Speech-to-Text transcription

Voice Options

Deepgram provides several AI voices with different characteristics:

• aura-asteria-en (Female, American)
• aura-athena-en (Female, American)
• aura-zeus-en (Male, American)
• aura-hera-en (Female, American)
• aura-orion-en (Male, American)

View the complete list in the voices.ts file or Deepgram's documentation.

New Features

• Speaker Selection: You can now specify a speaker voice when initializing the DeepgramVoice class. This allows for more personalized speech generation.

• Updated speak Method: The speak method now supports an optional speaker parameter in the options, allowing you to dynamically choose the voice for speech synthesis.