Choosing Plugins — Comparison & Recommendations¶

In a nutshell

OVOS ships dozens of interchangeable plugins for each job. That's useful, but it's easy to feel lost picking one. This page is the shortcut: for every plugin type it names a recommended default, then a small comparison table so you can see at a glance which option fits your situation (offline vs cloud, hardware budget, licence, maturity). Want the full config and every option? Each section links to its detailed catalog. New here? Start with the Plugins Index.

Don't want to choose? Use the recommended offline stack

Every ⭐ below is a sensible, fully-offline, on-device default. It forms a complete private stack: no cloud, no account. First install them:

pip install ovos-microphone-plugin-alsa ovos-vad-plugin-silero \
            ovos-stt-plugin-onnx-asr phoonnx

then drop this into ~/.config/mycroft/mycroft.conf:

{
  "listener": {
    "microphone": { "module": "ovos-microphone-plugin-alsa" },
    "VAD": { "module": "ovos-vad-plugin-silero" }
  },
  "stt": { "module": "ovos-stt-plugin-onnx-asr" },
  "tts": { "module": "ovos-tts-plugin-phoonnx" }
}

The wake word is not in the snippet because it already defaults to precise-onnx (hey_mycroft). See Wake word to change the phrase. Prefer one command? ovos-config autoconfigure -l en-us --offline writes the language + these defaults for you (it needs ovos-config already installed, and configures rather than installs plugins).

In a hurry with a specific constraint (low-power Pi, GPU accuracy, thin satellite, cloud OK)? Jump straight to Pick by scenario.

How to read the tables. ⭐ marks the recommended default. Maturity rates repository health (Proof-of-concept → Alpha → Beta → Stable → Mature), not how good the plugin is. A Beta default can be the right pick. Offline means it runs on-device with no network. Online calls a cloud service (separate terms). Hybrid talks to a server you can self-host. Pick the row whose "choose this if" matches you, then open the catalog page for its config.

Speech input¶

Microphone¶

Captures audio. See the Microphone catalog.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-microphone-plugin-alsa	Stable	offline	Linux/ALSA hardware — lowest latency, best performance
ovos-microphone-plugin-sounddevice	Stable	offline	Cross-platform (Linux/macOS/Windows) capture
ovos-microphone-plugin-pyaudio	Beta	offline	You want PortAudio bindings directly, no `speech_recognition` dep
ovos-microphone-plugin-files	Stable	offline	Automated testing — feed audio files instead of a live mic

VAD (Voice Activity Detection)¶

Detects when speech starts/stops. See the VAD catalog.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-vad-plugin-silero	Stable	offline	Best accuracy, esp. in noise; small neural model ships in the package (16 kHz)
ovos-vad-plugin-webrtcvad	Stable	offline	Lighter CPU-only fallback with tunable aggressiveness
ovos-vad-plugin-noise	Stable	offline	Lightest possible — energy threshold, no model at all

Wake word¶

Listens for the activation phrase. See the Wake word catalog.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-ww-plugin-precise-onnx	Beta	offline	The default `hey_mycroft` engine — accurate, lightweight always-on ONNX
ovos-ww-plugin-openWakeWord	Stable	offline	Better accuracy than Vosk from open pre-trained models, no training
ovos-ww-plugin-vosk	Stable	offline	Fastest setup for an arbitrary phrase, no model training (good for dev)
ovos-ww-plugin-wakewordlab	Alpha	offline	Very compact (~240 KB) models with a Silero pre-filter (install from source)
ovos-ww-plugin-wakeforge	Alpha	offline	Train a custom detector from a single phrase
ovos-ww-plugin-server	Alpha	hybrid	Thin satellite offloading detection to a self-hosted ww-server. Not available yet — repos still private

ovos-ww-plugin-precise-lite is deprecated. It is the TFLite predecessor of precise-onnx, kept working as a fallback but not a pick for new setups.

STT (Speech-to-Text)¶

Transcribes speech to text. See the STT catalog (16 plugins, top picks shown).

Plugin	Maturity	Runs	Choose this if
⭐ ovos-stt-plugin-onnx-asr	Beta	offline	Recommended default: ONNX Runtime, no PyTorch, int8 weights for low-RAM devices
ovos-stt-plugin-fasterwhisper	Stable	offline	Whisper accuracy via CTranslate2 int8/float16, CPU or GPU
ovos-stt-plugin-whispercpp	Stable	offline	Lightweight Whisper via whisper.cpp (tiny/small models) on modest CPUs
ovos-stt-plugin-nemo	Stable	offline	NVIDIA NeMo Conformer models, GPU strongly recommended
ovos-stt-plugin-vosk	Stable	offline	Long-established lightweight engine from a local model folder
ovos-stt-plugin-server	Stable	hybrid	Offload STT to an OVOS STT server. Set `urls` — unset, it sends audio to public community servers
ovos-stt-plugin-azure	Stable	online	No on-device compute budget; accept cloud STT (Microsoft terms)

Speech output¶

TTS (Text-to-Speech)¶

Turns replies into speech. See the TTS catalog (20 plugins, top picks shown).

Plugin	Maturity	Runs	Choose this if
⭐ ovos-tts-plugin-phoonnx	Stable	offline	Recommended default: small multilingual ONNX voices, auto model fetch
ovos-tts-plugin-coqui	Stable	offline	Higher-quality neural synthesis, heavier local compute
ovos-tts-plugin-espeakNG	Mature	offline	Very broad language coverage, tiny footprint, robotic voice
ovos-tts-plugin-pico	Mature	offline	Lightest possible offline voice on constrained hardware
ovos-tts-plugin-server	Mature	hybrid	Offload synthesis to an OVOS/Piper server. Set `host` — unset, it uses public community servers
ovos-tts-plugin-edge-tts	Stable	online	Free high-quality Microsoft Edge cloud voices, no subscription
ovos-tts-plugin-polly	Mature	online	A specific commercial cloud voice (Amazon Polly)

Licence watch

Most speech plugins are Apache-2.0, but a few differ. For example, espeakNG is GPL-3.0, and cloud engines (Azure, Polly, and others) add the vendor's separate terms. The catalog pages carry the exact licence per plugin.

G2P (Grapheme-to-Phoneme)¶

Converts text to phonemes to drive mouth-movement / viseme animation (e.g. a Mark 1 face). This is an alpha-quality, Mark 1-era capability. Most TTS voices don't emit phoneme timing, so a G2P plugin estimates it from the text. Only Mimic 1 provides phoneme timing natively. For any other voice the G2P plugin simulates the timing. If you don't drive a mouth/face, you don't need a G2P plugin at all. See the G2P catalog.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-tts-plugin-mimic	Beta	offline	The PyPI-published default — ships the `ovos-g2p-plugin-mimic` entry point, ARPA phonemes via the Mimic 1 engine

Language¶

Translation & Language Detection¶

Detect a text's language and translate. See the Translation catalog.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-translate-plugin-nllb	Stable	offline	Recommended offline translation (NLLB-200 via CTranslate2)
⭐ ovos-lang-detector-fasttext-plugin	Stable	offline	Recommended offline detection, pairs with NLLB
ovos-lang-detector-classics-plugin	Stable	offline	Offline detection voting across classic detectors
ovos-translate-server-plugin	Stable	hybrid	Self-/community-hosted translate server with public failover (talks to an `ovos-translate-server` instance)
ovos-google-translate-plugin	Stable	online	Free cloud translate+detect when coverage beats keeping text local

Media playback¶

Backend players for ovos-media. See the Media catalog. Note: the ovos-media service itself is alpha and not officially released. Stock installs play audio through ovos-audio (enable_old_audioservice: true, the default), so these backends only apply once you opt in to ovos-media. The maturity column below rates each backend plugin, not the service.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-media-plugin-mpv	Stable	offline	General-purpose local audio+video; works on both media backends
ovos-media-plugin-mplayer	Stable	offline	MPlayer as the local audio+video backend
ovos-media-plugin-ffplay	Stable	offline	Lightweight ffplay audio-only playback
ovos-media-plugin-vlc	Beta	offline	Headless VLC audio+video, ovos-media-ready package
ovos-media-plugin-spotify	Stable	online	Spotify Connect streaming
ovos-media-plugin-chromecast	Stable	online	Cast audio+video to a Chromecast on the network

Media & OCP¶

OCP Stream Extractors¶

Turn a URL or search request into a playable stream for OCP. See the OCP catalog for the full descriptions. None of the entries there state a maturity rating.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-ocp-youtube-plugin	not rated	online	Recommended default: resolves the most common "play X" request, YouTube/YouTube Music
ovos-ocp-files-plugin	not rated	offline	Playing local files (`file://` URIs), always available, no network
ovos-ocp-rss-plugin	not rated	online	Podcast/RSS feeds, always plays the newest episode
ovos-ocp-bandcamp-plugin	not rated	online	Bandcamp pages
ovos-ocp-news-plugin	not rated	online	A known spoken-news provider URL
ovos-ocp-m3u-plugin	not rated	online	Resolving a `.m3u`/`.pls` playlist URL
ovos-media-classifier	not rated	offline	Routes a request to the right `MediaProvider` by intent; deployed as a dependency of `ovos-ocp-pipeline-plugin`

Embeddings stores¶

Backends for EmbeddingsDB, the vector store an agent engine uses for retrieval or memory. See the Agent Plugins catalog. None of the entries there state a maturity rating.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-chromadb-embeddings-plugin	not rated	offline	Recommended default: local persistent client, no separate server to run
ovos-qdrant-embeddings-plugin	not rated	hybrid	Scaling to a self-hosted Qdrant server for a larger vector store
ovos-gguf-embeddings-plugin	not rated	offline	Already running ovos-gguf-plugin, reuse it for text embeddings too, no extra service

AI Agents & LLM Chat Backends¶

The "brain" behind a persona: a chat engine answers when no skill handles the utterance. See Agent Plugins, OpenAI Plugin, GGUF Plugin, and Personas for the full catalogs and config. None of those pages state a maturity rating for these plugins.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-gguf-plugin	not rated	offline	Recommended default: fully local GGUF model via `llama-cpp-python`, no account, no network, private
ovos-openai-plugin	not rated	online	Pointing at OpenAI or any OpenAI-compatible endpoint (cloud, Ollama, vLLM, LocalAI, `ovos-persona-server`)
ovos-messagebus-chat-plugin	not rated	offline	Answering through the existing skills/intent stack instead of an LLM
ovos-solver-rivescript-plugin / ovos-solver-aiml-plugin	not rated	offline	A lightweight scripted chatbot, no model weights at all. The repos put the words in a different order (`ovos-solver-plugin-rivescript`, `ovos-solver-plugin-aiml`), but the package and handler names are the ones shown here
ovos-wikipedia-plugin	not rated	online	Factual questions answered from Wikipedia
ovos-wolfram-alpha-plugin	not rated	online	Computational/factual questions (math, unit conversion, …)
ovos-ddg-plugin	not rated	online	General web-search-style answers via DuckDuckGo

Privacy of the demo persona

ovos-openai-plugin ships a Remote Llama demo persona pointed at a public third-party server. For privacy, point api_url at your own server, or use ovos-gguf-plugin instead.

Text & audio transformers¶

Hook into the text / audio / dialog / TTS stages. See the Transformers catalog.

Plugin	Maturity	Runs	Choose this if
⭐ ovos-utterance-normalizer	Stable	offline	Standard utterance cleanup before intent parsing (the default)
ovos-utterance-corrections-plugin	Stable	offline	Correct STT outputs for better intent matching
ovos-utterance-plugin-cancel	Stable	offline	Cancel an utterance ending in "nevermind that"
ovos-dialog-normalizer-plugin	Stable	offline	Normalize dialog text before TTS
ovos-bidirectional-translation-plugin	Stable	hybrid	Understand + speak in any language via paired transformers
ovos-audio-transformer-plugin-ggwave	Mature	offline	Data-over-sound encode/decode on the audio path

Pick by scenario¶

Fully offline / private (Raspberry Pi or mini-PC): the ⭐ stack: alsa mic, silero VAD, precise-onnx wake word, onnx-asr STT, phoonnx TTS. Everything runs on-device.
Lowest-power / tiny device: swap TTS to pico or espeakNG, VAD to webrtcvad, STT to vosk (small model). Accept lower accuracy for a smaller footprint.
Best accuracy, GPU available: STT fasterwhisper with use_cuda: true — what ovos-config autoconfigure --gpu configures for the languages that have a GPU recommendation, using whisper-large-v3-turbo where no dedicated finetune exists for the language (see Language Support Tables) — or nemo. TTS coqui. Still fully offline.
Thin satellite / shared backend: offload with the *-server plugins so the heavy models live on one machine. STT and TTS are available today (ovos-stt-plugin-server, ovos-tts-plugin-server — set urls/host, or they use public community servers); the wake-word one is not released yet, so keep wake-word detection on the satellite.
Cloud is acceptable (coverage/quality first): STT azure, TTS edge-tts/polly, translation google. Each adds the vendor's separate terms.

Maturity here is a repository-health signal, not a quality score. See the Maturity scale. For discovery/loading internals see the Plugin Manager. For head-to-head benchmark data see the Plugin Arena.

Read next: Plugin Arena · Microphone Plugins Related: Plugins Index · Maturity Scale · Plugin Manager