Piper (text to speech) 1.8.0 (piper-tts): our test log

Piper 1.8.0 synthesised 10 sentences (about 41 s of 22.05 kHz audio) on CPU in 1.4 to 1.5 s each including process start; 3.5 s uv install, 167 MB.

Result at a glance

What we found

• All 10 sentences synthesised with exit code 0; synthesis took 1.38 to 1.52 s per sentence, of which most is process and model start, so a long-running process would be faster per sentence (not measured).
• Sample rate was 22,050 Hz for this medium-quality voice.
• Round trip: the audio was clear enough that faster-whisper recovered it with a 2.4 percent word error rate in the best configuration (see the faster-whisper test).

What we ran, step by step

# Install with uv
uv venv -p 3.13 work/py/piper && VIRTUAL_ENV=work/py/piper uv pip install --no-cache -q piper-tts

# Download voice
python -m piper.download_voices en_US-lessac-medium

# Synthesise ten sentences
python -m piper -m en_US-lessac-medium -f s<N>.wav -- "<sentence>"

Raw output (excerpts)

0: 1.52 s for 5.12 s of audio: The agentic weekly newsletter covers new releases from open source projects every week.
1: 1.46 s for 3.32 s of audio: Please add seventeen and twenty five, and tell me the result.
2: 1.47 s for 3.68 s of audio: Local language models can run on a processor without a graphics card.
3: 1.45 s for 5.21 s of audio: The model context protocol lets an assistant call tools over standard input and output.

Gotchas and things that broke

• The Python package is GPL-3.0 licensed (OHF-Voice/piper1-gpl); check that licence before embedding it in a product.
• Voices are separate downloads; the package alone cannot speak.

What we did not test

• Voice quality as judged by a person (we only measured machine round-trip accuracy), other voices and languages, streaming, and GPU use.
• One voice only; a smoke test.

Reproduce it

Versions change. These commands install whatever is latest on the day you run them unless a version is pinned, so expect different results later. The version we measured is in the table above.

uv venv -p 3.13 .venv && VIRTUAL_ENV=.venv uv pip install piper-tts
.venv/bin/python -m piper.download_voices en_US-lessac-medium
.venv/bin/python -m piper -m en_US-lessac-medium -f out.wav -- "Hello from Piper."

How to read this test

Every number on this page was measured on this one machine, which also ran other jobs, so timings are indicative and vary run to run. A pass means our small task worked once or as stated, a fail means it did not work as we ran it; neither says how the tool behaves with a vendor's frontier model, on a GPU, or at scale. Where a tool could not run without an account or a GPU, we say so instead of guessing.

See the testing rig for the hardware, the local model and the rules we follow, and the test hub for every other test.

Related tests: voice, speech and image

• whisper.cpp: We could not run whisper.cpp: no prebuilt Linux release was available for download and this machine has no make, g++ or cmake to build it.
• ComfyUI: ComfyUI 0.42.0 installed in 17.42 s and its CPU-only server was ready in 30.1 s with 1014 node types, but we did not generate an image.
• faster-whisper (speech to text): faster-whisper 1.2.1 transcribed 41 s of speech at 2.4 percent word error with the base.en model in float32 (deterministic), but the int8 setting gave 2 to 60 percent on identical audio across runs.

All test logs · How we test

FAQ

Sources

• Piper repository (GPL fork) (fetched 2026-10-25)

Related

• All test logs
• Methodology and testing rig
• Study: a voice round trip with Piper and faster-whisper