¶ Configuring Deepgram in AVR
¶ Overview
Deepgram is an AI-powered speech platform providing Automatic Speech Recognition (ASR), Text-to-Speech (TTS) and Speech-to-Speech (STS) capabilities.
ASR Features:
- Real-time and batch transcription.
- High accuracy with advanced models like Nova-2.
- Support for over 36 languages and dialects.
- Low latency (under 300 ms for streaming).
- Options for diarization, keyword boosting, and filler word removal.
TTS Features:
- Natural-sounding, low-latency voice synthesis.
- Multiple voices and styles across supported languages.
- Configurable models, voices, and versions.
¶ How to Obtain a Deepgram API Key
- Sign up at Deepgram Console (free tier available with trial credits).
- Select your default project or create a new one.
- Navigate to Settings → API Keys.
- Click Create a New API Key.
- Provide a friendly name, choose appropriate permissions, and optionally set an expiration date.
- Click Create Key and copy the key immediately — it will not be shown again.
- Store the API key securely; do not commit it to public repositories.
¶ Deepgram ASR Setup
¶ Repository
- GitHub: avr-asr-deepgram
¶ Environment Variables
| Variable | Description | Example Value |
|---|---|---|
PORT |
Port where the ASR module listens | 6010 |
DEEPGRAM_API_KEY |
Your Deepgram API key | abc123xyz... |
SPEECH_RECOGNITION_LANGUAGE |
Target language code for transcription | en-US |
SPEECH_RECOGNITION_MODEL |
Deepgram model name (including version, if applicable) | nova-2-phonecall |
Example .env section:
DEEPGRAM_API_KEY=your_deepgram_api_key_here
SPEECH_RECOGNITION_LANGUAGE=en-US
SPEECH_RECOGNITION_MODEL=nova-2-phonecall
PORT=6010
Docker Compose Configuration:
avr-asr-deepgram:
image: agentvoiceresponse/avr-asr-deepgram
platform: linux/x86_64
container_name: avr-asr-deepgram
restart: always
environment:
- PORT=6010
- DEEPGRAM_API_KEY=$DEEPGRAM_API_KEY
- SPEECH_RECOGNITION_LANGUAGE=en-US
- SPEECH_RECOGNITION_MODEL=nova-2-phonecall
networks:
- avr
¶ Deepgram TTS Setup
¶ Repository
- GitHub: avr-tts-deepgram
¶ Environment Variables
| Variable | Description | Example Value |
|---|---|---|
| PORT | Port where the TTS module listens | 6011 |
| DEEPGRAM_API_KEY | Your Deepgram API key | abc123xyz... |
| DEEPGRAM_TTS_MODEL | Deepgram TTS model in format [model]-[voice]-[language]-[version] |
aura-arabella-en-xxhifi |
Example .env section:
DEEPGRAM_API_KEY=your_deepgram_api_key_here
DEEPGRAM_TTS_MODEL=aura-arabella-en-xxhifi
PORT=6011
Docker Compose Configuration:
avr-tts-deepgram:
image: agentvoiceresponse/avr-tts-deepgram
platform: linux/x86_64
container_name: avr-tts-deepgram
restart: always
environment:
- PORT=6011
- DEEPGRAM_API_KEY=$DEEPGRAM_API_KEY
- DEEPGRAM_TTS_MODEL=aura-arabella-en-xxhifi
networks:
- avr
¶ Deepgram STS Setup
¶ Repository
- GitHub: avr-sts-deepgram
¶ Environment Variables
| Variable | Description | Default / Example Value |
|---|---|---|
PORT |
Port where the STS module listens | 6033 |
DEEPGRAM_API_KEY |
Your Deepgram API key | abc123xyz... |
AGENT_PROMPT |
System prompt that defines the AI agent’s behavior | "You are a helpful assistant" |
DEEPGRAM_SAMPLE_RATE |
Audio sample rate | 8000 |
DEEPGRAM_ASR_MODEL |
Deepgram ASR model used for transcription | nova-3 |
DEEPGRAM_TTS_MODEL |
Deepgram TTS model used for speech generation | aura-2-thalia-en |
DEEPGRAM_GREETING |
Optional greeting message sent at the start of a session | "Hello! How can I help you?" |
OPENAI_MODEL |
OpenAI model for generating responses | gpt-4o-mini |
Example .env section:
DEEPGRAM_API_KEY=your_deepgram_api_key_here
AGENT_PROMPT=You are a helpful assistant
DEEPGRAM_SAMPLE_RATE=8000
DEEPGRAM_ASR_MODEL=nova-3
DEEPGRAM_TTS_MODEL=aura-2-thalia-en
DEEPGRAM_GREETING=Hello! How can I help you?
OPENAI_MODEL=gpt-4o-mini
PORT=6033
Docker Compose Configuration:
avr-sts-deepgram:
image: agentvoiceresponse/avr-sts-deepgram
platform: linux/x86_64
container_name: avr-sts-deepgram
restart: always
environment:
- PORT=6033
- DEEPGRAM_API_KEY=$DEEPGRAM_API_KEY
- AGENT_PROMPT=$AGENT_PROMPT
- DEEPGRAM_SAMPLE_RATE=8000
- DEEPGRAM_ASR_MODEL=nova-3
- DEEPGRAM_TTS_MODEL=aura-2-thalia-en
- DEEPGRAM_GREETING="Hello! How can I help you?"
- OPENAI_MODEL=gpt-4o-mini
networks:
- avr
¶ How AVR Uses Deepgram
ASR Module: AVR Core streams audio to Deepgram ASR, receives transcribed text, and passes it to the LLM.
TTS Module: AVR Core sends the LLM’s text response to Deepgram TTS, which returns audio to Asterisk or the calling application.
STS Module: The STS Module combines speech-to-text (STT/ASR) and text-to-speech (TTS) in a single flow.