Introducing Scribe v2

Concept 1: What is Scribe v2 and What Problem Does It Solve?

The Core Idea

Scribe v2 is a speech-to-text (transcription) model built specifically for handling large-scale, complex audio.

The Problem It Addresses

Previous transcription tools struggled with:

• Long recordings that drift in quality
• Extended silences or pauses
• Changes in tone or speaking style
• Diverse accents and speakers

Think of It This Way

Imagine trying to transcribe a 2-hour podcast with multiple guests, background noise, and technical jargon. Basic tools fail here. Scribe v2 is engineered precisely for this challenge.

Key Distinction

---

Concept 2: Keyterm Prompting — Context-Aware Transcription

The Core Idea

Standard transcription tools use fixed vocabulary lists — they always transcribe a word the same way regardless of context.

Keyterm Prompting is smarter. It uses the surrounding context of the transcript to decide when a specific term actually applies.

How It Works

1. You provide up to 100 words or phrases as keyterms
2. Scribe v2 reads the context around each spoken word
3. It intelligently decides whether to apply your keyterm or not

Why This Matters

Without context awareness:

"The patient needs an MRI scan" → might transcribe as "em-ar-eye"

With Keyterm Prompting:

The model recognizes the medical context and correctly applies "MRI"

Best Use Cases

• ✅ Technical/scientific terminology
• ✅ Brand names
• ✅ Industry-specific language (legal, medical, finance)

---

Concept 3: Built-in Entity Detection with Timestamps

The Core Idea

Beyond just transcribing words, Scribe v2 can automatically identify and locate sensitive information within your audio.

What Is an "Entity"?

An entity is a specific category of meaningful information, such as:

• 🔒 PII — names, addresses, phone numbers
• 🏥 Health data — diagnoses, medications
• 💳 Payment details — credit card numbers, bank info

What Makes This Powerful

Scribe v2 supports up to 56 detection categories and provides:

• The detected entity itself
• Its exact timestamp in the audio

Real-World Application

A compliance team processing recorded customer calls can automatically flag and redact credit card numbers spoken aloud — without manually reviewing hours of audio.

The Workflow It Enables

Audio Recording → Transcription → Entity Detection → Flag/Redact/Review

---

Concept 4: Automatic Multi-Language Transcription

The Core Idea

Real-world audio doesn't always stay in one language. Scribe v2 handles multiple languages within a single audio file automatically.

How Traditional Tools Handle This

• You must manually segment the audio by language
• Or run separate transcription jobs per language
• This is slow, error-prone, and expensive at scale

How Scribe v2 Handles This

1. You upload one audio file — no pre-processing needed
2. The model detects each language as it appears
3. It transcribes each segment correctly in its respective language

Example Scenario

A multinational company records a meeting where speakers switch between English, Spanish, and French. Scribe v2 handles all three in a single pass.

---

Concept 5: Production-Grade Features for Enterprise Use

The Core Idea

Scribe v2 isn't just accurate — it's built for real production environments with features that developers and enterprises actually need.

Breaking Down Each Feature

🎙️ Smart Speaker Diarization

• Automatically labels who is speaking and when
• Produces clean, readable transcripts for multi-speaker recordings

⏱️ Word-Level Timestamps

• Every single word gets a precise time marker
• Enables accurate subtitle alignment and interactive transcript experiences

🔊 Dynamic Audio Tagging

• Detects non-speech events like:
  • Laughter
  • Footsteps
  • Applause
• Adds context beyond just words

🔐 Enterprise Compliance

Scribe v2 meets major security and privacy standards:

Plus: EU and India data residency and zero retention mode (your data isn't stored).

---

Summary: The Full Picture

SCRIBE V2
│
├── Core Strength → Accurate batch transcription at scale
│
├── Keyterm Prompting → Context-aware custom vocabulary
│
├── Entity Detection → Automatic sensitive data identification + timestamps
│
├── Multi-Language → Single file, automatic language switching
│
└── Production Features
        ├── Speaker diarization
        ├── Word-level timestamps
        ├── Audio event tagging
        └── Enterprise compliance

The One-Line Takeaway

Scribe v2 transforms raw, complex, real-world audio into accurate, structured, compliant, and actionable transcripts — at scale.