What changed in AI today

• The break tag works on all ElevenLabs models except Eleven v3 and v4, with pauses up to three seconds.
• Phoneme tags for custom English pronunciation work only on Flash v2 and Turbo v2.
• Eleven v3 and v4 drop SSML entirely in favor of bracketed Audio Tags such as [whispering] and [sighs].
• WebSocket streams need enable_ssml_parsing turned on, and tags should be sent whole to avoid added latency.

Why it matters for marketers: Content teams producing voice scripts must check the model first, since break and phoneme support differs and v3 and v4 need Audio Tags instead.

• Upload a video in ElevenCreative and the model generates original music plus sound effects placed at matching timestamps.
• Audio-as-reference option lets the new score follow the genre, tempo, or instrumentation of existing audio.
• Sound effects appear as separate layers in the Music Video Editor for independent review and adjustment.
• Studio API adds add_sfx, separate_tracks, and use_audio_as_reference; separate_tracks=true returns music and SFX as separate files in a ZIP.

Why it matters for marketers: Content teams can add scored music and timed sound effects to short-form video in one tool, and mix stems separately via the API.

• Watermark detectors give a yes or no answer, but only for the provider that embedded the watermark.
• ElevenLabs' Audio Detector finds only ElevenLabs-generated watermarks, not those from other companies.
• Statistical speech classifiers return a confidence score and can cover audio from any provider, though accuracy varies.
• The free AI Speech Classifier analyzes only the first minute and is unreliable on Eleven v3 audio.

Why it matters for marketers: Content and fraud teams should treat watermark checks as one layer in a multi-tool stack, since no single detector covers all voice providers.

• Scribe v2 supports verbatim and clean transcripts, with verbatim as the default setting in Playground and the API.
• Multichannel transcription, using a separate microphone per speaker, labels speakers accurately even when cross-talk occurs.
• Keyterm prompting allows up to 1,000 terms per transcript on Scribe v2, or 50 on Scribe v2 Realtime.
• Transcripts can be exported as .txt, .docx, .pdf, .json, .html, .srt, or .vtt, and Scribe v2 supports over 90 languages.

Why it matters for marketers: Content teams gain a checklist for choosing verbatim or clean output and microphone setups that reduce speaker attribution errors in quotes.

• The Web Speech API sends microphone audio to a browser-chosen cloud engine, so transcription results vary across browsers.
• Developers can't control voice models, and the API lacks reliable custom vocabulary support, such as for the brand name ElevenLabs.
• The API has no speaker diarization and no server-side streaming or audio file uploads, only live microphone input.
• The source recommends ElevenLabs Scribe v2 Realtime for commercial products, citing latency below 150 ms and 90+ languages.

Why it matters for marketers: Teams building voice features in web apps should weigh browser-dependent results and missing features before relying on native speech recognition in production.

• Brand Kit stores colors, fonts, logos, reference images and usage rules as one workspace object.
• Users tag the kit with @ in Image & Video, Flows, or the ElevenCreative agent to shape outputs.
• Agencies and multi-brand teams can create a kit per client and switch between them.
• ElevenLabs says it is working to extend kits to brand voice and sound; no timing given.

Why it matters for marketers: Marketing and content teams can stop re-describing brand identity in every prompt and share one set of guidelines across the whole team.

• VAD outputs a yes-or-no decision on speech in each frame, typically 10-30 milliseconds, without outputting words.
• The process runs as a five-stage loop: capture, framing, feature extraction, scoring from 0 to 1, and state decision.
• Traditional VAD uses energy thresholds and can false-trigger on loud noise; neural and semantic AI variants handle noise and context better.
• Endpointing, not VAD, decides whether a speaker has finished a thought; the two are complementary technologies.

Why it matters for marketers: Content teams using voice agents or transcription should benchmark VAD on noisy, real-world calls rather than clean recordings, since false positives and negatives affect results.

• Scribe v2 Realtime reports 150 ms latency from audio input to text output, for live captioning during meetings.
• Scribe v2 batch supports up to 1,000 keyword terms, while Scribe v2 Realtime supports up to 50 key terms.
• Batch transcription runs asynchronously over REST, with a webhook alerting developers when the transcript is ready.
• Models support 90+ languages, and ElevenLabs cites Zero Retention Mode for enterprise privacy needs.

Why it matters for marketers: Content and product teams building meeting tools can compare real-time versus batch transcription and choose between Scribe v2 Realtime and Scribe v2 for their workflows.

• Five stages: data capture, transcription, AI analysis, insights, and CRM integration, per the guide.
• Scribe v2 Realtime streams transcription with around 150 ms latency for in-call coaching or compliance alerts.
• Speaker diarization covers up to 32 speakers, and transcription spans more than 90 languages.
• The guide says the platform supports SOC 2, HIPAA, GDPR, and EU data residency.

Why it matters for marketers: Content and marketing teams get a checklist of pipeline stages and compliance requirements to weigh when scoping call-analysis projects.

• Scribe returns tagged, timestamped word, spacing, and audio_event tokens in one API call, with no forced-alignment pass.
• Audio events such as laughter and applause are tagged when tag_audio_events is enabled, making reactions searchable for highlights.
• Multichannel transcription supports up to 5 channels, each transcribed independently with deterministic speaker IDs.
• Output exports to JSON, SRT, and VTT, and live WebSocket streams can include timestamps with include_timestamps=true.

Why it matters for marketers: Content teams can build searchable archives, captions, and clips from one transcription call instead of a separate alignment pipeline.