Configure the Non-Verbal detection rule

The Non-Verbal detection rule flags Session items whose audio is mostly non-speech — grunts, sighs, laughter, breath, silence. It runs upstream of transcription: audio classified as non-verbal skips transcription and is auto-tagged with a Property of your choice.

Use this to clean up takes where the actor didn’t actually deliver a line, before they pollute your Session items table.

Open the panel

1. Open the Session you want to configure.
2. Open the Auto-rules panel from the Session toolbar.
3. Switch to the Non-Verbal detection tab.

The panel carries a master Enabled / Disabled toggle at the top.

The three sections

1. When to Apply

Same pattern as the Mismatch rule — add filters to scope this rule to Sessions that carry specific Properties at the Session level. Leave empty to apply to every Session in your account.

2. Speech Detection Sensitivity

The core knob: a slider from 0.00 to 1.00 controlling how selective the rule is. VoiceQC labels the ranges so you don’t have to guess:

Threshold	Label	Effect
0.00–0.30	Very Sensitive	Almost everything is sent to transcription. Few items tagged non-verbal.
0.31–0.50	Sensitive	Most audio transcribed.
0.51–0.70	Balanced	The middle. 0.60–0.70 is the recommended range.
0.71–0.90	Strict	Quieter clips and partial takes get flagged non-verbal.
0.91–1.00	Most Strict	Only confidently-spoken audio passes through.

The rule of thumb VoiceQC surfaces under the slider:

Lower = more audio sent to transcription. Higher = more audio tagged as non-verbal.

You can drag the slider or type a value into the numeric input next to it. The recommended starting point is 0.70 (Balanced).

Iterate, don't overthink

Set the threshold, click Re-evaluate Session, and look at what got flagged. If too much real dialogue is tagged non-verbal, lower the threshold. If too much silence/breath made it through to transcription, raise it. Two or three passes usually dial it in.

3. Assigned Property

The Property VoiceQC applies when audio is classified as non-verbal. Click Change Property to pick from your Property library, or define a new entry on the spot. A common choice is Status / Non-verbal or Quality / No take.

Save and re-evaluate

1. After changing the threshold or the Property, the Save Changes button activates in the footer. Reset discards your edits.

2. Click Save Changes.

3. Click Re-evaluate Session to apply the updated rule to existing items. A toast tracks progress and shows how many Properties were added and removed when the run completes.

The rule also applies automatically to new items as they’re processed. You only need to re-evaluate when you change the rule itself, not when new audio arrives.

Mismatch vs Non-Verbal

These two rules look at different things:

• Non-Verbal runs on the audio before transcription. It catches takes that shouldn’t be transcribed in the first place.
• Mismatch runs after transcription. It catches takes that were transcribed fine but don’t match the expected script.

Use both together. Non-Verbal cleans the input; Mismatch QCs the output.

For deeper guidance on when each is the right tool, see Mismatch vs Non-Verbal: choosing the right rule.

See also

• Configure a Mismatch rule — the other built-in Auto-rule.
• Auto-rule configuration — concise reference for both rules side-by-side.
• Glossary: Auto-rule, Non-Verbal rule