TTS (Text-to-Speech) Resource

The TTS resource synthesizes speech from text, producing an audio file that can be accessed by downstream resources via the ttsOutput() expression function. It supports both cloud (online) and local (offline) synthesis engines and can be used as a primary resource or as an inline resource inside before / after blocks.

Basic Usage

apiVersion: kdeps.io/v1
kind: Resource

metadata:
  actionId: speakResponse
  name: Speak Response

run:
  tts:
    text: "Hello, welcome to KDeps!"
    mode: offline
    offline:
      engine: espeak

---

Configuration Options

Option	Type	Description
text	string	Required. Text to synthesize. Supports expressions.
mode	string	Required. online or offline.
language	string	Optional BCP-47 language code (e.g. en-US).
voice	string	Voice identifier (provider/engine-specific).
speed	float	Speech rate multiplier. 1.0 is normal speed.
outputFormat	string	Audio format: mp3 (default), wav, or ogg.
outputFile	string	Explicit output path. If omitted, a path under /tmp/kdeps-tts/ is auto-generated.
online	object	Cloud provider config (required when mode: online).
offline	object	Local engine config (required when mode: offline).

Online Provider Configuration

tts:
  mode: online
  online:
    provider: openai-tts     # Required: provider name
    apiKey: sk-...           # API authentication key
    region: us-east-1        # Region (AWS Polly / Azure only)
    subscriptionKey: ...     # Azure Cognitive Services subscription key

Field	Description
provider	One of: openai-tts, google-tts, elevenlabs, aws-polly, azure-tts.
apiKey	Authentication credential for the provider.
region	AWS region or Azure region (e.g. eastus).
subscriptionKey	Azure-only: Cognitive Services subscription key.

Offline Engine Configuration

tts:
  mode: offline
  offline:
    engine: piper            # Required: engine name
    model: en_US-lessac-medium  # Model name or path (piper / coqui-tts)

Field	Description
engine	One of: piper, espeak, festival, coqui-tts.
model	Model name or file path (used by piper and coqui-tts).

---

Online Providers

OpenAI TTS

Calls the OpenAI /v1/audio/speech endpoint.

tts:
  text: "Hello from OpenAI TTS"
  mode: online
  voice: alloy         # alloy | echo | fable | onyx | nova | shimmer
  outputFormat: mp3
  online:
    provider: openai-tts
    apiKey: sk-...

Google Cloud Text-to-Speech

Calls the Google Cloud TTS REST API.

tts:
  text: "Hello from Google Cloud"
  mode: online
  language: en-US
  voice: en-US-Standard-A
  speed: 1.0
  online:
    provider: google-tts
    apiKey: AIza...

ElevenLabs

Calls the ElevenLabs /v1/text-to-speech endpoint.

tts:
  text: "Hello from ElevenLabs"
  mode: online
  voice: "21m00Tcm4TlvDq8ikWAM"    # ElevenLabs voice ID
  online:
    provider: elevenlabs
    apiKey: xi-...

AWS Polly

Note: AWS Polly requires SigV4 request signing. Use the exec resource to call the AWS CLI instead:

run:
  exec:
    command: "aws polly synthesize-speech --output-format mp3 --voice-id Joanna --text 'Hello' /tmp/speech.mp3"

Azure Cognitive Services TTS

Calls the Azure TTS REST API.

tts:
  text: "Hello from Azure"
  mode: online
  language: en-US
  voice: en-US-JennyNeural
  online:
    provider: azure-tts
    region: eastus
    subscriptionKey: ...

---

Offline Engines

Offline engines run locally without any network calls. They must be installed on the host system.

Piper

High-quality neural TTS. GitHub: rhasspy/piper

# Install
pip install piper-tts
# Or download binary from GitHub releases

tts:
  text: "Hello from Piper"
  mode: offline
  offline:
    engine: piper
    model: en_US-lessac-medium    # Download from https://huggingface.co/rhasspy/piper-voices

eSpeak-NG

Fast, lightweight speech synthesizer. eSpeak NG

# Install
apt install espeak-ng          # Debian/Ubuntu
brew install espeak            # macOS

tts:
  text: "Hello from eSpeak"
  mode: offline
  voice: en             # eSpeak voice identifier
  speed: 1.2            # Words per minute multiplier
  offline:
    engine: espeak

Festival

Classic speech synthesis system. Festival Speech Synthesis System

# Install
apt install festival            # Debian/Ubuntu
brew install festival           # macOS

tts:
  text: "Hello from Festival"
  mode: offline
  offline:
    engine: festival

Coqui TTS

Open-source deep-learning TTS. Coqui TTS

# Install
pip install TTS

tts:
  text: "Hello from Coqui"
  mode: offline
  offline:
    engine: coqui-tts
    model: tts_models/en/ljspeech/tacotron2-DDC

---

Accessing the Output File

After the TTS resource runs, the path to the synthesized audio file is available via:

• ttsOutput() — expression function in any resource evaluated after TTS runs
• get("ttsOutput") — unified get() helper
• input("ttsOutput") or input("tts") — input() helper

resources:
  - metadata:
      actionId: speakThenRespond
    run:
      tts:
        text: "Here is your answer"
        mode: offline
        offline:
          engine: espeak

  - metadata:
      actionId: sendAudioFile
    run:
      apiResponse:
        success: true
        response:
          audioPath: "{{ttsOutput()}}"

---

TTS as an Inline Resource

TTS can run before or after the main resource action using the before / after inline blocks:

run:
  before:
    - tts:
        text: "Processing your request, please wait…"
        mode: offline
        offline:
          engine: espeak

  chat:
    model: llama3
    prompt: "{{input()}}"

  after:
    - tts:
        text: "Processing complete."
        mode: offline
        offline:
          engine: espeak

---

Using Expressions in text

The text field supports KDeps expressions:

tts:
  text: "Hello {{get(\"name\")}}, your score is {{get(\"score\")}} points."
  mode: offline
  offline:
    engine: espeak

---

Example: Voice Assistant Response

Full example combining input transcription with TTS output:

# workflow.yaml
settings:
  input:
    source: audio
    audio:
      device: hw:0,0
    activation:
      phrase: "hey kdeps"
      mode: offline
      offline:
        engine: faster-whisper
        model: small
    transcriber:
      mode: offline
      output: text
      offline:
        engine: faster-whisper
        model: small

# resources/respond.yaml
run:
  chat:
    model: llama3
    prompt: "{{inputTranscript()}}"

# resources/speak.yaml
run:
  tts:
    text: "{{get(\"respond\")}}"
    mode: offline
    offline:
      engine: piper
      model: en_US-lessac-medium