Live Voice Translator

Live Transcription Controls:

Speaker mic (phones)

Ready to listen

Mic input (raw — before any browser processing) 🎤 Volume

Processed input — what the ASR actually hears, after ticks Note 1 & 2 below (noise suppression + auto-gain, echo cancellation)

Auto-detect language Specify language:🗣 I speak: Streaming mode (Live ASR API) ; untick means use token-expensive non-Live model

(Auto picks the best transcription engine for the language you choose — Fun-ASR for all, except Filipino → Gemini. You can also pick an engine manually.)

Economy standby — don't send detected silence to the ASR engine; except for the trailing s after silence detected, then After s of quiet, mark the sentence finished — Fun-ASR method: for s (Fun-ASR only — Gemini always receives zero silence; the duration applies to the audio-send modes; live-editable)

Trigger phrase mode: reset sentence speech Silence sensitivity: (for economy gate, auto-stop & level bar; Auto tracks the noise floor +18. The higher the value, the more easily sound counts as "silence") Auto-stop listening after seconds of silence Note 1: Noise suppression (Removes steady background noise) + auto-gain (Normalizes volume, boosts quiet/far voices, tames loud spikes) (browser DSP cleans the mic signal — cuts steady background noise and boosts quiet/far voices before the ASR hears them) Note 2: Echo cancellation — ignore this device's own audio (requires Note 1) (same speakerphone tech as Zoom/Teams calls — subtracts your app's own TTS playback from the mic so the ASR doesn't transcribe it; works on Chrome/Edge/Android, weaker on iOS)

Live Transcription:

Translation Controls:

Translate to: Translation Engine:

(Auto-picks best engine for each language — Qwen-flash LLM for all, except Minnanhua → Gemini.)

During Live-Transcribe: Live-Translate immediately Live-Audio immediately

Audio Controls:

Audio TTS Engine:

(Auto-picks best engine for each language — CosyVoice TTS: Minnanhua | Microsoft Voice TTS: Filipino | Browser voice if available, otherwise Microsoft voice.)

Voice speed: 1.0x Keep CosyVoice engine warm (Pinging Alibaba server 1 char per 4 min to keep engine active and immediate — tick box takes effect immediately) Show phrase grouping (underlined units = word-groups, wider gap = bigger boundary)

Translation Result: 🔒 read-only

Auto-play audio after translation

Marker Parsing: 🔒 read-only

API Performance Metrics (for program developer):