Voice ID

Your Voice Is Your Password

Read one sentence three times. Sathi builds a mathematical fingerprint of your voice — no recording stored, just numbers. From that moment, only your voice unlocks your world. Reject strangers. Verify you. All on-device. Delete the profile anytime with one tap.

Set Up Voice ID All Features
How It Works

Enrollment → Verification

Enrollment — 3 Voice Samples🎙️Sample 1"My voice is my identity.This app should respond🎙️Sample 2"My voice is my identity.This app should respond🎙️Sample 3"My voice is my identity.This app should respondECAPA-TDNN Speaker Embedding ModelExtracts a fixed-length float32 vector representing your unique voice characteristicsWeighted Average of 3 Embeddingsnew_avg = (old_avg × n + new_sample) / (n + 1)🔒Encrypted Realm StorageNo raw audio stored — only embedding vectorVerification — Every ConversationNew Voice RecordingExtract embedding in real timeCosine Similaritydot(a,b) / (‖a‖ × ‖b‖)✅Accept≥ 0.75 similarity🔄Retry0.65 – 0.75🚫Reject< 0.65 similarity
Features

Voice Security Without Friction

3-Sample Enrollment

Read one sentence three times. Sathi extracts a unique voiceprint from each sample, averages them into a single speaker embedding, and stores it — encrypted — on your device. The raw audio is discarded immediately.

Real-Time Verification

Every time you speak, Sathi extracts a fresh embedding and compares it with your stored profile using cosine similarity. Accept at 0.75+, retry between 0.65–0.75, reject below 0.65. All in under a second.

No Raw Audio Stored

Only the mathematical embedding — a fixed-length float array — lives on your device. There is no recording file, no audio snippet, nothing a human could listen to. The vector is meaningless without the model.

Re-Enroll or Delete Anytime

Changed your mind? Reset your voice profile with one tap. Want to re-enroll? Record three new samples. Your voice profile belongs entirely to you — on your device, under your control.

My Voice Mode

Once enrolled, enable My Voice Mode and Sathi adapts its speech style — pitch, rate, and energy — to complement your natural voice pattern. Calm voices get gentle responses. Expressive voices get livelier ones.

Works During Calls Too

Voice ID isn't just for chat. When you use Sathi's voice calling features, speaker verification runs on the call audio to confirm it is you before executing sensitive commands. Extra security, zero extra steps.

Enrollment in 60 Seconds

1

Read the prompt

"My voice is my identity. This app should respond only to me." — spoken clearly, 1.5 to 15 seconds.

2

Repeat twice more

Each sample is validated for text similarity (Jaccard + coverage ≥ 0.55) and duration. If it's too short or too different, you re-record.

3

Profile stored

Three embeddings are averaged into one robust voiceprint. Stored in encrypted Realm. Raw audio deleted. Done.

Only Your Voice. Nobody Else's.

Set up Voice ID and make Sathi truly yours.

Download Free Back to Overview