Choose ASR + translation + TTS, click Start, and wait for microphone audio to be received. Say a short sentence, pause, and wait for the translation and audio. Repeat, then Stop. Each row below is one finalized utterance.
All values are milliseconds (1,000 ms = 1 second). These measure work on the server. They do not measure microphone upload or the delay until you hear sound. Do not add them together as end-to-end latency.
End of this utterance’s audio at the server → final transcript. Includes waiting for the end-of-speech pause.
Translation job starts → translated text ready. Includes waiting and retries.
TTS job starts → first audio bytes ready at the server. This is not when your speakers start playing.
Full output ready: —. Includes completion and any ordered delivery wait.
Click a sentence to inspect it above. Median = middle result; p95 = 95% of samples were this fast or faster. Small samples give a rough estimate. “—” means not measured yet or unavailable, never zero. Merged utterances share one translation/TTS job. First-audio timing is available only when the provider reports streaming audio.
| Utterance / original text | ASR final | Translation | TTS first audio | TTS full output |
|---|---|---|---|---|
| Start a run and speak to collect measurements. | ||||
ASR delays run from audio end at the server to the result, including endpointing. Open Latency test for per-sentence results and run summaries.
Paste text and translate to test translation alone. The time shown runs from the backend starting the request until translated text is ready, including waiting and retries.
Paste text and generate speech to test TTS alone. First audio is the server wait for the first audio bytes; full output is the wait for all audio. Neither measures when your speakers play it.