@pipedog/deepgram

Deepgram services for pipedog: streaming speech-to-text (DeepgramSTTService), WebSocket and HTTP text-to-speech (DeepgramTTSService, DeepgramHttpTTSService), and Flux, Deepgram's turn-taking STT (DeepgramFluxSTTService) and its TTS (DeepgramFluxTTSService). It talks to the API with fetch and WebSocket; no Deepgram SDK is needed.

A port of Pipecat's pipecat.services.deepgram. Subpaths mirror upstream: @pipedog/deepgram/stt, /tts, /flux/stt, /flux/tts.

Install

bun add @pipedog/core @pipedog/asyncio @pipedog/deepgram

No peer dependencies.

Usage

import { DeepgramSTTService } from "@pipedog/deepgram/stt";
import { DeepgramTTSService } from "@pipedog/deepgram/tts";

const apiKey = process.env.DEEPGRAM_API_KEY ?? "";

const stt = new DeepgramSTTService({ apiKey });
const tts = new DeepgramTTSService({ apiKey, sampleRate: 24000 });
// new Pipeline([transport.input(), stt, userAggregator, llm, tts, transport.output(), ...])

DeepgramFluxSTTService (@pipedog/deepgram/flux/stt) detects turns itself: use it with no VAD analyzer and nothing runs locally.

Runtimes

Bun, Node.js 22+, Deno and Cloudflare Workers.

Not every upstream Deepgram module is ported yet (for example the SageMaker variants); see porting status.

Links

• Docs: turn detection, voice bot guide
• pipedog README
• License: BSD 2-Clause, a port of Pipecat (Copyright (c) 2024–2026 Daily). See LICENSE.