Benchmark

Measure speech recognition, voice synthesis, and full conversation turns with the same local engines used by the app.

Speech to Text benchmark

Run every selected STT endpoint against the same reference clip and compare speed, accuracy, model, and compute device.

Pick a library voice to use its WAV and reference transcript.
STT engines No engines loaded.
Ready
EngineModelDeviceTimeAccuracyOutput
Run a benchmark to see results.

Run benchmark

Pick a backend and voice, set run count, then measure synthesis latency and real-time factor (RTF). RTF < 1.0 means the backend generates faster than real-time.

Batch benchmark

Benchmark multiple voices in one run. Pre-populated from your active My Voices — or reload from the backend. Results are sorted fastest first and saved to History.

Uses the sample text and Runs value from the single-voice form above. Check one voice, a few voices, or Select all, then run them together.

History

Last 50 benchmark sessions saved in your browser. Each row is one run session — click the backend/voice to pre-fill the form above.

No benchmark history yet. Run a benchmark above to start tracking.

Conversation turn benchmark

Runs the same STT → LLM → TTS pipeline as Conversation Playground and records turn latency.

Ready

Run a turn to see transcript, reply, and timing.

Latency
STT
—
LLM first token
—
LLM total
—
TTS
—
Total
—
Turn history
No turns yet.