Voice Samples
One clean 2‑minute sample powers every voice/video call—and improves audio/video training.
What it is
Capture a single, clear ~30‑second recording that your Delphi uses on every voice or video call. The same sample also helps train how your Mind handles audio/video content.
Best practice: Record in a silent room with a high‑quality mic, and upload one consistent recording—don’t mix clips of different quality.
Where: Studio → Voice
Quick answers
Where do I record/upload? Go to Voice on the left console. If you’ve never recorded, the recorder opens automatically; otherwise click ＋ (top‑right).
Why am I not able to call? Ensure your Access Group limit isn't set to 0, a valid sample is saved, and the browser has mic permissions.
How do I enable training on audio files? Upload clean audio/video under Mind → Files / YouTube / Podcasts. Your voice sample helps match your speech; other speakers are stored for context but not credited to you.
How to add a voice recording
Option 1 — Upload File (WAV/MP3)
Click ＋ → Upload Samples (We strongly encourage 1 great sample, rather than multiple).
Select your WAV/MP3 (~30 sec) file.
Click Upload again.
Option 2 — Start Recording (in browser)
Click ＋ → Start Recording and grant mic access.
Speak clearly for ~30 sec—steady tone, single speaker, no background noise.
Press Stop (▢) → Save.
Turn it on & tune it
Click settings to adjust Stability, Similarity, Speed. Use Voice Playground to test without changing global settings; click Apply settings to my Delphi only when satisfied.
Go to Access Groups to enable and disable voice by clicking the three dots on each group
Best practices
Silent space: Turn off fans/AC and notifications; ensure no other voices.
One strong sample: A single, steady 30-second take beats many mixed clips.
Hold tone & volume: Keep mouth 6–8" from the mic; speak evenly.
One language per sample: Mixed‑language recordings degrade quality.
Gear: USB mic is fine; XLR + interface (e.g., AT‑2020/RØDE NT1 + Focusrite) is better. Avoid Bluetooth headsets and Zoom/phone recordings.
Live test: Make a real call; it’s the truest preview of what users hear.
Sample script (≈120 seconds)
“Hi, I’m [Name]. This sample helps Delphi learn how I naturally speak so calls feel personal and clear. I work on [area/expertise], and people usually come to me for [topics]. I aim to be [tone—e.g., warm and practical], focusing on [outcomes]. Here’s how I might explain something: when you’re starting with [topic], begin by [step 1], then [step 2]. If you’re unsure, I’ll ask a clarifying question and point you to examples. If you asked about [frequent question], I’d say: [short 1–2 line answer]. Thanks for listening—this should capture my cadence, pacing, and energy.”
Troubleshooting
Quality sounds off: Re‑record in a quieter room; stay 6–8" from mic; reduce echo; tweak Similarity/Stability.
Too fast/slow: Adjust Speed in voice settings.
Upload stuck: Keep files near 30 seconds, use WAV/MP3, refresh and retry.
Pro Voice (optional)
If your current voice is decent but you want studio‑grade realism, consider Pro Voice. It trains a dedicated model on your recordings (10–30 minutes recommended). See the Pro Voice page for setup.
Pre‑launch checklist
Last updated