Whisper speech-to-text hardware

Home Assistant Whisper Hardware Requirements

Whisper is the compute-heavy part of many fully local Home Assistant voice pipelines. Home Assistant currently notes roughly eight seconds on Raspberry Pi 4 versus under a second on an Intel NUC for incoming commands, illustrating why server class matters.

See live hardware Size my workload

Quick answer

What this voice workload needs

Intel N100 or equivalent is the current Home Assistant minimum recommendation for Whisper Base in a fully local Voice Preview Edition setup. Languages that need Whisper Small or Large can require substantially more powerful hardware, including GPU acceleration.

Live Amazon hardware

Current products that fit this decision

These listings come from this sprint's dedicated Amazon catalogue. Barebones PCs, laptops, GPU enclosures, accessories, ambiguous voice-device listings and incomplete systems are excluded by the normalizer.

Checking the dedicated Home Assistant Voice & Local AI catalogue…

Buying decision

Match compute to speech and LLM load

Use the smallest Whisper model that gives acceptable recognition for your language. Upgrade hardware when transcription latency, larger model requirements or multiple simultaneous requests justify it—not because a voice tutorial says every installation needs a discrete GPU.

Interactive decision tool

Whisper Hardware Sizer

Size the voice and local-AI workload before comparing products. Results are planning tiers, not benchmark guarantees. Verify current Home Assistant voice support and Ollama hardware support for the exact system.

Compatibility checklist

Four checks before you purchase

Choose focused or open-ended STT first

Speech-to-Phrase and Whisper have very different compute needs.

Treat LLM memory separately

Model size, quantization, context and concurrency drive RAM/VRAM.

Keep Piper in perspective

Text-to-speech is usually lighter than Whisper or Ollama.

Measure end-to-end latency

Microphone, network, STT, LLM and TTS all affect response time.

Whisper is the expensive stage in a full local pipeline

Home Assistant’s own local-voice guide makes the contrast clear: Whisper attempts open-ended transcription and therefore costs far more compute than focused Speech-to-Phrase. The server has to turn arbitrary audio into text before Home Assistant can act. When a local voice system feels sluggish, Whisper is often the first stage to profile rather than Piper or the automation engine.

Use the N100 recommendation as a minimum planning point

For Voice Preview Edition users who want fully local Whisper Base, Home Assistant currently recommends an Intel N100-equivalent processor or better. That recommendation is useful because it anchors buying decisions in the application instead of generic CPU charts. It still leaves room for differences in thermal limits, background containers and language. A tiny N100 box is a starting tier, not a universal solution for every Whisper model.

Base, Small, Medium and Large are different hardware projects

Moving up the Whisper model ladder changes latency and memory demands enough that the correct machine can change. Base can be practical on efficient x86 hardware. Small often rewards a stronger modern CPU. Medium and Large are more likely to justify GPU assistance when conversational response time matters. Select the model for recognition quality first, then size the host around that choice rather than buying hardware before testing language quality.

Language can matter more than room count

An English household that gets good results from Whisper Base can have a lighter hardware requirement than a one-room installation whose language needs a larger model. Home Assistant explicitly warns that model requirements vary by language quality. Before buying, test representative names, rooms and commands in the language that will actually be spoken. Recognition accuracy is part of performance; a fast transcript that misunderstands devices is not a successful voice server.

Concurrent transcription is a separate sizing dimension

A single Whisper request may complete quickly while two or three simultaneous requests queue behind one another. Multi-room households should therefore distinguish total satellite count from expected concurrent speech. If only one person normally speaks at a time, eight endpoints do not automatically require eight times the compute. If several family members can issue commands together, extra cores or GPU throughput can improve perceived responsiveness.

Detection of silence and audio quality affect effective workload

Whisper only receives useful audio after capture and wake-word stages have done their jobs. Poor microphones, reverberation or noisy rooms can create longer or repeated utterances that feel like compute problems. A better endpoint may reduce retries more than a faster CPU. Benchmark the server with the same satellite, distance and acoustic environment that will be used in production.

GPU acceleration is about latency, not prestige

A supported GPU becomes compelling when the chosen Whisper implementation and model can actually use it and CPU latency is no longer acceptable. Do not select a graphics card from gaming rankings alone. VRAM, driver support, container access, idle power and chassis fit matter. For a dedicated speech server, a modest efficient GPU can be more appropriate than a high-end card whose compute remains mostly idle.

Whisper and Ollama should not silently compete for one GPU

A combined voice-and-LLM server may ask the same GPU to transcribe speech and generate a response. That can be efficient when workloads are small, but it can also cause latency spikes if VRAM is tight or tasks overlap. Monitor GPU memory and scheduling under a real voice conversation. If contention is obvious, use a larger-memory card or separate the speech and LLM services across different devices.

CPU-only Whisper remains attractive for simple deployments

GPU acceleration is not required to build a useful fully local voice assistant. N100, N305 and mobile Core/Ryzen processors can provide a compact CPU-based path with no extra graphics-card power or driver stack. CPU-only designs are easiest to maintain when Base or Small meets the language requirement. Their main limitation is response time as model size and concurrency rise.

RAM and SSD capacity are supporting requirements

Whisper itself does not usually require the enormous memory associated with large LLMs, but the complete Home Assistant host still needs room for the operating system, integrations, databases and other voice services. Sixteen gigabytes is a comfortable modern baseline for a dedicated x86 speech server; 32GB makes more sense when Ollama or many containers share it. A 512GB–1TB SSD is normally sufficient unless a large model library also lives there.

Measure transcription separately from the rest of Assist

End-to-end voice timing can hide where seconds are spent. Record the interval from completed utterance to available transcript first. Then measure intent/LLM processing and Piper separately. This prevents buying a new Whisper server because a local LLM is slow, or blaming Ollama for an oversized speech model. Repeat the measurement after model, language or satellite changes.

Upgrade path by symptom

If Base is accurate but slightly slow, a faster CPU may be enough. If the language needs Small or Large and response time becomes unacceptable, GPU-assisted speech processing is more likely to pay off. If multiple rooms create queues, prioritize concurrency. If recognition errors dominate, fix microphones or model choice first. Hardware upgrades should respond to the observed failure mode rather than a generic “best Whisper PC” list.

Questions people ask

Home Assistant Whisper Hardware Requirements questions

Can Home Assistant voice run fully locally?

Yes. Home Assistant documents fully local pipelines using local speech-to-text and Piper text-to-speech, with no requirement to send spoken commands to a cloud service.

What is the lightest local speech-to-text option?

Speech-to-Phrase is designed for focused home-control commands and can run very quickly on low-powered Home Assistant hardware.

What CPU does Home Assistant recommend for Whisper?

For fully local Voice Preview Edition use with Whisper Base, Home Assistant currently recommends at least an Intel N100 or equivalent processor.

Does Piper need a GPU?

Usually no. Piper is optimized for modest hardware and is much lighter than open-ended Whisper or a local LLM.

Does Home Assistant run Ollama itself?

The Home Assistant Ollama integration connects to an external Ollama server, which may be on the same machine or another computer reachable over the network.

How much context does Home Assistant use with Ollama?

Home Assistant currently defaults its Ollama integration to an 8K context window. Increasing context can improve capacity for larger homes but raises memory use.

How many entities should I expose to a local LLM?

Home Assistant currently recommends exposing fewer than 25 entities when experimenting with local LLM control, especially because smaller models can make more mistakes.

Is NVIDIA required for Ollama?

No. Ollama documents NVIDIA, supported AMD/ROCm, Apple Metal and experimental Vulkan paths. Compatibility depends on the exact hardware and operating system.

Should I put Ollama on a separate server?

Separate it when GPU power, memory pressure, experimentation or noise would compromise the always-on Home Assistant host. Small models can stay consolidated when simplicity matters more.

What should I measure after installation?

Measure end-to-end voice latency, Whisper transcription time, LLM first-token time, RAM/VRAM use, network delay and behavior when more than one request arrives.

Official references and methodology

Verify current Home Assistant and Ollama support

Cloudzat separates speech capture, speech-to-text, intent/LLM processing and text-to-speech because each stage has different hardware requirements. Current Home Assistant voice/Ollama documentation and current Ollama hardware/context documentation are the technical baseline. Live products come from this sprint’s dedicated Amazon catalogue; calculator results are planning tiers rather than guaranteed latency benchmarks.

As an Amazon Associate, Cloudzat may earn from qualifying purchases. Prices, seller terms, exact configurations and Home Assistant or Ollama hardware support can change.