Dating voice DM with verbal off-platform contact
Transcript flags it before delivery.
Daniel, 26
Active now
0:14
Voice note · transcribed
“…just text me on WhatsApp, +1 555-0142, we can skip the app”
Off-platform contact
Moderate spoken content with audio transcription
Convert voice notes, audio messages, podcasts, and live audio rooms to text and moderate the transcript with Lasso's four-layer moderation pipeline.
How audio transcription works in Lasso
1Transcription + ML
Audio extracted from voice notes, audio rooms, and audio uploads. Transcribed per segment.
ML classifiers run on each segment, e.g. hate, threats, PII, profanity.
See Lasso in actionLIVEAudio room · auto-transcribing
12:03Jordanyeah I think the patch fixed it
12:04Rikoyou clowns are tr*sh, every one of you—
12:04Marawhoa, not cool
Hate speech: 96%
Transcription + ML
Scoring
Segment · Riko 12:04
"you clowns are tr*sh…"Hate 96%
Also detected
Profanity 88% · Harassment 82%
Language
English · auto-detected
Top confidence96%
180ms · liveForwarded to rules
2Custom rules
Transcript content runs through your custom rules. Word lists, topic restrictions, brand-specific triggers.
Easily configured in our dashboard.
See Lasso in actionLIVEAudio room
12:04Rikoyou clowns are tr*sh, every one of you—
Rule Input
01Hate speech · 96% · block
02Profanity · 88% · block
03Slur list · matched · block
Custom Rules
Match
Rule · Audio rooms
Block hate speech & slur-list matches in live voice
Hate speech · 96% > 80%
Slur list · exact match
!
Action
Speaker muted · removed
3AI Moderator
Ambiguous transcript content reaches the AI Moderator. Intent and context evaluated.
Spoken slang, irony, coded language assessed.
See Lasso in actionVOICE NOTEDating DM · transcribed 0:14
0:02Danielhey, really enjoyed talking with you
0:09Danieljust text me on WhatsApp, +1 555-0142, we can skip the app
Off-platform · solicitation
AI Moderator
Reasoning
1
Transcribe
Voice note → text, per segment
2
Analyze
Handoff to WhatsApp + phone number spoken
3
Context
First voice DM, new match, dating
4
Classify
Off-platform solicitation
×
Verdict
Blocked before delivery
4Human review
Borderline transcribed content reaches moderators with everything they need.
The moment, the words, the AI's read on it.
See Lasso in actionVOICE NOTEDating DM · transcribed 0:08
0:05Danielhaha you could always just slide me your number, no pressure though
Queued for review
AI summary · borderline
Off-platform intent 71% · playful tone · below auto-action threshold
Human Review
Pending
Off-platform intent
71% probabilityBorderline
Tone
Playful · ambiguous
Context
First voice DM · new match
Confidence71% · below threshold
✓ Approve
× Remove
Decision retrains the AI Moderator
Three capabilities included in audio transcription
Spoken content in voice notes, audio messages, and podcasts converted to text. Segmented by utterance with timestamp per segment.
Voice note · Daniel
transcribing · 0:14
0:02hey, really enjoyed talking with you
0:06we should hang out sometime
0:09just text me on WhatsApp, +1 555-0142
Segment 03 · 0:09
“just text me on WhatsApp, +1 555-0142”
utterance · timestamped14
segments
each timestamped
each timestamped
Language-aware transcription. 18 languages supported. Language detected automatically from the audio.
Real-time transcripts for live audio. Transcribed segment by segment as the session runs. Flagged violations reach the queue without waiting for the session to end.
Voice note · Daniel
transcribing · 0:14
0:02hey, really enjoyed talking with you
0:06we should hang out sometime
0:09just text me on WhatsApp, +1 555-0142
Segment 03 · 0:09
“just text me on WhatsApp, +1 555-0142”
utterance · timestamped14
segments
each timestamped
each timestamped
Lasso: next-gen AI content moderation
99.9%
On autopilot, and getting smarter every day.
Three layers of AI handle the volume, your rules, and the grey areas. Your team only sees what needs them, and every decision they make improves the system.
Keep users safe without driving them away.
Customizable moderation that lets you find the right balance between safety and user experience. So you protect your community without suppressing the culture that makes it worth joining.
Complexity removed from content moderation.
One API. Clear dashboards. A moderation pipeline built around one-click actions and the right context, right where you need it.
★★★★★
4.9
Highest rated in content moderation on G2.
Every platform has spoken content. Audio transcription moderates it.
Dating
Voice notes and audio prompts carry contact info and solicitation.
- Voice notes reciting WhatsApp or Telegram handles to move matches off platform.
- Spoken phone numbers and payment-app handles hidden in audio profile prompts.
- Voice DMs delivering verbal solicitation, harassment, or unsolicited explicit content.
Audio communities & social
Live audio rooms and voice comments carry spoken abuse at broadcast scale.
- Hate speech spoken aloud in live audio rooms reaching broadcast audiences.
- Voice comments under posts containing verbal harassment and personal attacks.
- Audio replies on short-form video used to deliver verbal abuse at scale.
Podcasting & creator platforms
Long-form spoken content carries compliance and privacy risk most platforms can't review.
- Festival presenters and hosts saying slurs or threats on live audio feeds.
- Conference panellists or callers verbally revealing attendees' personal data on air.
- On-air guests reciting phone numbers or home addresses during live broadcasts.
FAQs
Audio transcription, answered
Lasso pulls spoken content from voice notes, audio rooms, and uploads, then transcribes it segment by segment. Each segment passes through the four-layer pipeline: ML classifiers score it, your custom rules apply, the AI Moderator weighs intent and context on anything ambiguous, and edge cases reach human reviewers with the moment, the words, and the AI's read on it. This automates about 99.9% of moderation decisions.
The four-layer pipeline automates around 99.9% of moderation decisions, and anything uncertain goes to human review with full context. ML classifiers return a confidence score for each segment, and the AI Moderator resolves spoken slang, irony, and coded language that rules alone miss. Accuracy depends on audio quality, language, and how much background noise is present.
Manual review means a person listens to every voice note, room, or podcast, which does not scale once spoken content grows. Lasso transcribes and screens the audio automatically and sends only the genuine edge cases to your team, each with the timestamp and the AI's reasoning attached. Your moderators spend their time on the calls that actually need human judgment.
Audio transcription runs through the same Lasso API and dashboard as the rest of your moderation. You connect your audio source, set your rules in the dashboard, and flagged segments land in the queue you already use. Live audio rooms are transcribed in real time so violations surface during the session rather than after it ends.
Audio transcription pulls spoken content from voice notes, audio messages, audio rooms, and podcasts and converts it to text. The transcript is run through content moderation to detect harmful speech.
Many component models run concurrently to evaluate vocal tone, volume, pitch, context, and emotion across seven categories: Adult Language and Sexual Vulgarity, Audio Assault, Gender and Sexual Hate Speech, Harassment and Bullying, Racial and Cultural Hate Speech, Violent Speech, and Other Harmful Speech.
Yes. Audio rooms and live voice broadcasts are transcribed segment by segment in real time. Violations are queued during the session.
18 languages for transcription and moderation, with the language detected automatically from the audio.
Both run on Modulate. Audio transcription handles asynchronous audio such as voice notes, podcasts, and audio rooms. Voice moderation handles real-time live voice chat in gaming and voice calls.
Moderate what users say
Lasso's audio transcription catches harmful speech in voice notes, audio rooms, and podcast uploads. One pipeline, one API.
Book a demo
Lasso Moderation
© 2026. All rights reserved.
