Kalpa for Pipecat

Kalpa text-to-speech services for Pipecat, maintained by Kalpa Labs, the company providing the Kalpa API.

This community integration provides KalpaTTSService for WebSocket streaming and KalpaHttpTTSService for HTTP streaming. See the Kalpa speech documentation for available voices, models, and languages.

Demo

Watch the Pipecat + Kalpa TTS 1 demo. The browser conversation includes English and Hindi speech, a spoken interruption, and timestamped Pipecat events.

Installation

Requires Python 3.11 or later and Pipecat 1.13.x. Automated compatibility tests pass with Pipecat 1.13.0 on Python 3.12.

uv add pipecat-kalpa

Usage

Get an API key from the Kalpa dashboard and set KALPA_API_KEY in your environment. See authentication for details.

import os

from pipecat.pipeline.pipeline import Pipeline
from pipecat_kalpa import KalpaTTSService

tts = KalpaTTSService(api_key=os.environ["KALPA_API_KEY"])

# Add Kalpa to an existing voice pipeline.
pipeline = Pipeline(
    [
        transport.input(),
        stt,
        user_aggregator,
        llm,
        tts,
        transport.output(),
        assistant_aggregator,
    ]
)

transport, stt, llm, and the context aggregators are the other components of your application. A complete runnable example is provided below.

Use KalpaTTSService to send a reply's text as it arrives over one WebSocket connection. For HTTP streaming, replace it with KalpaHttpTTSService(api_key=os.environ["KALPA_API_KEY"]).

Configuration

Both services require api_key. Pass voice and model overrides with settings=KalpaTTSService.Settings(...) or settings=KalpaHttpTTSService.Settings(...):

tts = KalpaTTSService(
    api_key=os.environ["KALPA_API_KEY"],
    settings=KalpaTTSService.Settings(
        voice="18c26f49-9f9c-400c-bec9-7467f28e1db5",
        model="kalpa-tts-1",
    ),
)

Voice and model changes can be sent during a session with TTSUpdateSettingsFrame. The WebSocket service reconnects to apply them.

The services produce 24 kHz mono, 16-bit PCM audio. Pipecat's output transport resamples it when your pipeline uses a different output rate. Do not pass sample_rate to these services.

Run the voice example

The example uses Deepgram for speech recognition, OpenAI for generating replies, and Kalpa for speech output. It connects to your browser over WebRTC.

git clone https://github.com/kalpalabs/pipecat-kalpa.git
cd pipecat-kalpa
uv sync --extra examples
cp .env.example .env

Set KALPA_API_KEY, DEEPGRAM_API_KEY, and OPENAI_API_KEY in .env, then run:

uv run --extra examples python examples/voice.py --transport webrtc

Open the address printed by the runner, enable microphone access, and connect. Speak while the assistant is talking to interrupt its reply.

Set KALPA_TTS_TRANSPORT=http to run the same example with HTTP streaming:

KALPA_TTS_TRANSPORT=http uv run --extra examples python examples/voice.py --transport webrtc

KALPA_VOICE_ID optionally overrides the example's voice.

Development

uv sync
uv run pytest
uv run ruff check .
uv run ruff format --check .
uv build

Tests use local fake API servers and do not need API keys. Report issues on the integration repository.

License

BSD-2-Clause. See LICENSE.